Paper deep dive
Protein Crystallization Segmentation and Classification Using Subordinate Color Channel in Fluorescence Microscopy Images
Truong X. Tran, Marc L. Pusey, Ramazan S. Aygun
Intelligence
Status: not_run | Model: - | Prompt: - | Confidence: 0%
Entities (0)
Relation Signals (0)
No relation signals yet.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:The accuracy of detecting protein crystals for fluorescence microscopy images is very critical for high throughput and automated systems. Although the trace fluorescent labeling method could highlight protein crystals, reflection and emission from the fluorescence dye is not always due to crystal regions. Therefore, the analysis of the peak wavelength in the emission spectra of a fluorophore may not always yield effective results. In this paper, we show that using the subordinate color intensity corresponding to longer wavelengths than the peak wavelength of the emission spectra could improve the accuracy of protein crystal detection. Hence, we have built a segmentation method based on the percentile intensity of the subordinate color for trace fluorescently labeled (TFL'd) protein crystallization trial images. Compared to using the dominant color channel, our segmentation method on subordinate color channel was able to reduce the misclassification rate of likely-leads or crystals as non-crystals by the percentage of from 9.71% to 2.02% depending on the classifier. Similarly, the accuracy of classifiers were increased by the percentage of from 1.77% to 5.53%. Our method reached around 94% accuracy while keeping misclassification of likely-leads and crystals as non-crystals below 1%. Moreover, to evaluate the generalizability of our method, we have conducted new wet lab experiments on two proteins, Concanavalin A (Con A) and Ab inorganic pyrophosphate (AbIPPase), and the misclassification rate was below 1%. Our experiments show that using the subordinate channel may be more helpful for TFL'd protein trial image classification.
Tags
Links
- Source: https://arxiv.org/abs/2610.15286v1
- Canonical: https://arxiv.org/abs/2610.15286v1
Trouble viewing inline? Open PDF directly â
Full Text
60,906 characters extracted from source content.
Expand or collapse full text
Noname manuscript No. (will be inserted by the editor) Protein Crystallization Segmentation and Classification Using Subordinate Color Channel in Fluorescence Microscopy Images Truong X. Tran·Marc L. Pusey·Ramazan S. Aygun AbstractThe accuracy of detecting protein crystals for fluorescence microscopy images is very critical for high throughput and automated systems. Although the trace fluorescent labeling method could highlight pro- tein crystals, reflection and emission from the fluores- cence dye is not always due to crystal regions. There- fore, the analysis of the peak wavelength in the emis- sion spectra of a fluorophore may not always yield effec- tive results. In this paper, we show that using the sub- ordinate color intensity corresponding to longer wave- lengths than the peak wavelength of the emission spec- tra could improve the accuracy of protein crystal detec- tion. Hence, we have built a segmentation method based on the percentile intensity of the subordinate color for trace fluorescently labeled (TFLâd) protein crystalliza- tion trial images. Compared to using the dominant color channel, our segmentation method on subordinate color channel was able to reduce the misclassification rate of likely-leads or crystals as non-crystals by the percent- age of from 9.71% to 2.02% depending on the classifier. Similarly, the accuracy of classifiers were increased by the percentage of from 1.77% to 5.53%. Our method reached around 94% accuracy while keeping misclassifi- cation of likely-leads and crystals as non-crystals below 1%. Moreover, to evaluate the generalizability of our method, we have conducted new wet lab experiments on two proteins, Concanavalin A (Con A) and Ab in- organic pyrophosphate (AbIPPase), and the misclassi- fication rate was below 1%. Our experiments show that Truong X. Tran and Ramazan S. Aygun Data Media Lab, Computer Science Department, The University of Alabama in Huntsville, USA. E-mail: txt0002,aygunr@uah.edu Marc L. Pusey XpressGenes, Inc, Huntsville, Alabama, USA. Chemistry Departement, The University of Alabama in Huntsville, USA. E-mail: mlp0041@uah.edu using the subordinate channel may be more helpful for TFLâd protein trial image classification. KeywordsFluorescence Image Analysis·Protein Crystallization·Image Segmentation·Classification 1 Introduction High-throughput systems could handle both setting up experiments and evaluating outcomes to aid crystallo- graphers in conducting a large number of experiments. Image acquisition, analysis, and classification are ma- jor components of evaluating thousands of trial images on a daily basis. Trace fluorescent labeling (TFL) uti- lizes fluorescence dye to identify crystals that would have high intensity when exposed to light at the corre- sponding wavelength. This technique allows to acquire a clear view of wells in plates thus enabling to build a standalone imaging and classification system for an- alyzing plates [6, 12, 13, 15]. Fig. 1 shows images of sample wells that were taken under white light and flu- orescence illumination. Image analysis techniques typically focus on the emis- sion spectra of the fluorescence dye to identify crystals and the phases of crystallization. Nevertheless, reagents in cocktail solutions may also reflect light. Fluorescence dye molecules may float in the solution until they at- tach to a protein crystal, and these may complicate the imagery analysis of crystallization process. Traditional techniques naturally focus on wavelengths around the excitation/emission wavelengths of the fluorescence dye and have to deal with reflections and emissions from non-crystal areas in the solution. Although reflection or emission is possible from non-crystal areas in solutions at dominant wavelengths, emission or reflection is likely to be lower at farther wavelengths from the dominant wavelength. As the fluorescence dyes attach to protein arXiv:2610.15286v1 [cs.CV] 11 Oct 2026 2Truong X. Tran et al. Fig. 1White light (top) and fluorescent images of non- crystal (left) and crystal (right) sample wells. crystals, their emission at farther wavelengths will also be significant due to their density and this could help to separate crystal regions from non-crystal regions. In this paper, we analyze whether processing based on subordinate channel corresponding to farther wave- lengths than the peak wavelength of the emission spec- tra could provide better results or not than methods relying on the dominant color channel (based on the peak wave length). In our previous work, we have uti- lized subordinate channel while building and evaluat- ing a new classification technique, the else-tree classifier [20]. In [19], we have built a framework for determining the best parameter set while classifying protein crystal images. Using this framework, in this paper we explain our segmentation method based on subordinate channel that could improve the accuracy of classifiers while re- ducing the misclassification rates compared to using the dominant color channel. We analyze percentile based thresholding methods to compare the performance of using features based on the subordinate color channel and dominant color channel. After the features are ex- tracted, we use a number of classifiers to build a learn- ing model and show how those features obtained from the corresponding channels affect the accuracy. Com- pared to our work in [19], this paper adds more details on the analysis and experimental results of using domi- nant and subordinate color channels. Moreover, to show the generalizability of our method, new wet-lab experi- Image Analysis Image Acquisition Screening Optimization -- Crystallization Screening Fig. 2Basic analysis pipeline of protein crystallization (refer to [19]). ments were conducted for Concavanlin A (Con A) and Ab inorganic pyrophosphate (AbIPPase) proteins, and we show that it is possible to reduce the misclassifica- tion rate of crystals as non-crystals below 1% using the subordinate color channel. This paper is organized as follows. The following section explains the related work on classification of protein crystallization trial images. Section I provides background on TFL, explains the data set used building our learning model, reviews the relevant thresholding techniques used in previous research, introduces the re- gion features extraction of the trial images, and presents our classification model building framework. Section IV covers our method including the concept of subordinate color, the analysis of TFL image intensity, and the sub- ordinate percentile intensity thresholding method. Ex- periments and results are discussed in Section V. Dis- cussion of experimental results is provided in Section VI. The last section concludes our paper. 2 Related Work As shown in Fig. 2, there are four major steps in the analysis of protein crystallization experiments include crystallization screening, image acquisition, image anal- ysis, and screening optimization. In the screening stage, the protein solution is prepared and dropped into a well-plate. The plates are scanned in the image acquisi- tion stage and then the captured images are stored in a database. During the image analysis stage, the images are classified or given a numerical score indicating the phase of crystallization. The relevant screens can be se- lected and optimized for getting higher quality crystals. There has been significant work on image segmenta- tion and classification to increase the quality of protein crystallization analysis. Since high-throughput screen- ing systems are automated systems, it is critical to re- duce the possibility of missing crystallization outcomes. In 2002, Spraggon et al. classified the result of protein crystallization experiments based on analyzing texture features and edge detection of trial images [18]. They used a self-organizing neural network for the classifi- cation model. The model accuracies were from 47% to 82% based on different categories (from clear drop to mountable crystal). In 2004, Zhu et al. analyzed the trial images using geometric and texture features for classifying crystal and non-crystal screening results using a support vector machine and automatic deci- sion tree model. They reported the lowest false-negative rate of 9.6% while the best false positive rate is 14.6% [21]. Bern et al. (2004) analyzed the gradient and ge- ometric features of the in-focus region in the image of the sample solution [1] and classified images into five Protein Crystallization Segmentation and Classification Using Subordinate Color Channel3 classes (âemptyâ, âclearâ, âprecipitateâ, âmicrocrystal hitâ and âcrystalâ) using the decision tree algorithm. The testing results yielded about 12% false negatives and 14% false positives. In 2006, Pan et al. extracted the texture and Gabor wavelet decomposition from the trial images [10]. Pan et al. also used the support vector machine to classify three categories of clear drop, pre- cipitate, and crystal results. They obtained a low false negative rate of under 4%. However, the false-positive rate was about 40%. In addition, there were many studies on analyzing a larger set of features or using the image itself as the in- put for training classification models. In 2008, Po et al. detected the regions of interest and applied multi-scale Laplacian pyramid filters and histogram analysis on the trial images [11]. Po et al. combined the genetic algo- rithms with neural network classification models. Their method produced 99% true negative and 88% true pos- itive rate. In 2010, Cumbaa et al. obtained the set of basic image statistics, energy, Euler number, Radon- Laplacian, Sobel-edge, microcrystal, and GLCM fea- tures [3]. They trained a random forest classification model to classify three categories of crystal-bearing im- ages, precipitate images, and clear drops. The classi- fication accuracy per category is 80%, 89%, and 98% respectively. As an application of deep learning, Bruno et al. (2018) [2] presented a machine learning model that could correctly recognize 94% of test images while categorizing images into four categories as crystals, pre- cipitate, clear, and other outcomes. Their deep neural network was trained on half a million crystallization trial images. In our works, we have analyzed and classified TFLâd protein crystallization trial images for our automated fluorescence microscopy system. Sigdel et al. (2013) [13] obtained the intensity and region features to classify three classes (non-crystal, likely lead and crystal) of the trial images. We got 88% overall accuracy with only 1.2% misclassifying crystal as non-crystal rate when us- ing using multiple perceptron neural network (MLP) and max-class ensemble models. In 2014, Sigdel et al. [14] used decision tree model to classify four crystal image categories (needles, small crystals, large crys- tals, and other crystals) based on geometric features. In 2017, Sigdel et al. (2017) [16] evaluated several com- binations of features sets with different classification models. We also applied feature reduction and selec- tion techniques to improve classification performance. In our analysis, we got 96% accuracy with 87% sensi- tivity on three classes of non-crystals, likely leads and crystals using a large feature set consists of intensity features, region features of Otsuâs thresholding, green percentile thresholding G90 and G99, histogram fea- tures, and graph features. Most of the research studies on categorizing pro- tein crystallization trial images captured under white light. These studies used different feature extraction techniques and classification models on varying num- ber of protein crystallization categories. Although it is difficult to compare the performance of these methods, in general, the researchers claim an accuracy around 90% with misclassification rate of crystals around 10%. We should note that the research studies usually re- port their results based on the light source the im- ages are captured. Previously, we have built the else- tree classifier [20] and a framework that determines the best features, methods, parameters, and classifiers for fluorescence images [16, 19]. Recently, Klijn and Hub- buch (2019) [8] present their work on combining the end points and time-dependent features using 3 differ- ent light sources for classifying the protein phases. The combination of different light sources have a better bal- ance accuracy result than a simple complex classifica- tion models. However, in the literature, for fluorescence microscopy, we have not seen the comparison of classifi- cation based on dominant color wavelength and subor- dinate color channel. Hence, it is important to analyze how subordinate color channel processing impacts clas- sification accuracy. 3 Trace Fluorescent Labeling and Image Classification Protein crystallization experiments utilize characteris- tics of a specific protein and chemical solution to pro- duce successful screening results. Robotic high-throughput systems help to carry out many experiments with fast trial image capturing capability. The analysis of the trial images is essential for a better understanding and observing the screening result. Altogether, the goal is to increase the success of the crystal growth process. This section reviews the basics of our protein crystallization process, our TFL2756 trial image data set, our previ- ous image segmentation method, the summary of our region features extraction, and the general framework to build and evaluate our classification model. 3.1 Trace Fluorescent Labeling Protein Crystallization Analysis In our wet lab, we applied the Trace Fluorescent Label- ing (TFL) method to rapidly reveal the appearance of the crystal in a trial solution. Thus the crystals of la- beled protein have a high concentration of fluorescence 4Truong X. Tran et al. Trace Fluorescent Labeling Crystallization Screening Fluorescent imaging Image analysis Optimization trials Fig. 3Stages of protein crystallization analysis using TFL (refer to [19],©2017 IEEE.). dye and they can be visualized by fluorescence imaging [6, 12]. Fig. 3 shows the major steps in our protein crys- tallization experiments. We first label the protein with fluorescence dye. Then, we prepare the sample solution during the screening phase. After the preparation, the sample plates are carefully sealed and stored in an in- cubator. The plates are scanned periodically. Normally the first scan starts right after setting up the sample plates and follows by either a daily or weekly period. For capturing the image of each trial well of the plates, we place each of them in the dark chamber of a fluores- cence microscope. Then, the microscope automatically scans all wells of the plate. In the image analysis phase, each trial image is visualized, segmented, and classi- fied with scores. The score is an evaluation outcome of the experiment with respect to the crystallization phase ranging from a clear drop non-crystal to a big 3D crys- tal. There are many other phases of crystallization as well. All of the image analysis processes are computer- supported. The experimenter is also allowed to review the trial plates to verify the computer analyzed results. The analysis results are used for the later optimization experiments to optimize the leading crystalline condi- tions. We use our custom Crystal X2 fluorescence micro- scope [13] as the main instrument for imaging and anal- ysis the crystallization sample plates. The layout of Crystal X2 is presented in Fig. 4. The microscope is composed of an illumination component, optical com- ponent, automated control mechanism, and computer- aided software. The microscope uses LED illuminators to minimize the changing of temperature when captur- ing the sample images. The optical component contains a low pass excitation filter (Omega XF1080), a dichroic mirror (Omega XF2012), a high pass emission filter (Omega XF3021), and an imaging camera. The con- trol mechanism automatically slides the plate in the X and Y direction to locate the wells based on the plate camera X axis Y axis Z axix excitation lter emission llter dichroic mirror excitation light plate holder plate Fig. 4Layout of fluorescence microscope system. configuration[12]. When a well is located, the software controls the camera to capture the trial image. During image capturing, the light (LED) goes through the excitation filter and is reflected by the dichroic mir- ror to focus on the sample well. The emitted fluores- cence light from the well sample solution is collected by the same dichroic mirror, then gone through the emis- sion filter toward the camera lens. The software controls the camera to capture the trial image from the well. Fi- nally, all trial images are stored in the database together with other relevant experiment information [12, 13]. 3.2 TFL2756 data set In our research, we gathered a data set of 2756 trial images from different TFL experiments. The crystallo- grapher expert manually reviewed and classified each image into three categories based on Hamptonâs scale [16]. The categories and frequencies of images in this study are displayed in Fig. 5. The bar chart shows the number of images in non-crystal, likely-lead, and crystal categories. Around 60% of the images are non- crystal. The images which do not contain clear crystals but whose conditions may lead to crystals are labeled as likely-leads. The use of an intermediate class (likely- lead) may reduce the costly mistake of missing crystal without requiring to review all non-crystals [13]. Fig. 6 shows sample images of the TFL2756 data. Each image has dimensions of 320Ă240, 32-bit depth, and 96 dpi resolution. Protein Crystallization Segmentation and Classification Using Subordinate Color Channel5 The distrubution of TFL2756 trial image data set 1600 675 481 non-crystal likely-leadcrystal 0 200 400 600 800 1000 1200 1400 1600 1800 Fig. 5Distribution of TFL2756 data set. 3.3 TFLâd Image Segmentation based on Dominant Color Wavelength The intensity-based analysis is suitable for TFL images since protein crystals have a higher intensity than the background. Our TFL imaging uses green LED light (530nm) as the excitation source, thus the dominant color channel in the trial images is green. Therefore, in our previous work, we used a green percentile intensity thresholding method to segment the images [13]. We have used green percentile thresholding while develop- ing our decision-tree based binarization method [4] and supervised thresholding method to choose the best bi- narization method [5]. Moreover, we have conducted extensive studies using this threshold for generating ge- ometric features [14] and comprehensive evaluation of feature sets [16]. In our studies, G90, G95, G97, G99, and G100 are five thresholding that generated proper results. For instance, in the case of G90 thresholding, a threshold value T G90 is computed as the 90 th percentile intensity of the green channel. In this case, 90% of the pixels in the image have green intensity less than the threshold T G90 . The green percentile intensity thresh- olding also uses a grayscale (t min ) threshold to require the pixels of interest to have some minimal intensity if crystallization occurs. For image binarization, a pixel is considered as fore- ground if its overall gray intensity is greater than t min and green intensity greater than T G90 . On the other hand, the maximum green intensity thresholding, G100, uses T G100 - the maximum green intensity among all pixels in the image. In this case, the foreground con- stitutes all pixels with the green component intensity equal to T G100 and gray level larger than t min . In the rest of the paper, to indicate that the green channel is the dominant color channel, we will add prefixDom in front of green percentile thresholding. For example, Fig. 6Non-crystal (a-d), likely-lead (e-h), and crystal (i-l) sample images. DomG90 is the same asG90 but it indicates that green is the dominant color channel. 3.4 Region Features Extraction There is a trend to increase the number of image fea- tures to improve the classification performance. How- ever, the use of irrelevant features may reduce the per- formance of some classifiers. The comparison and anal- ysis of features for protein crystallization were present in one of our work [16]. In this paper, we use connected region features from the original trial image and gen- eral binary image features from its binarized image. For the binary image features, we focus on the intensity of the image since the crystals have the highest emission compared to other precipitates. For the region features, we focus on the shape and size of the large blobs. For each binary image, 7 features are extracted: the ratio of foreground to image size, the ratio of con- vex hull area to image size, foreground average inten- sity, standard deviation (stdev) of foreground intensity, background average intensity, standard deviation for background intensity, and the number of connected re- gions (blobs). We used connected labeling to segment the blobs in the binary image. When the regions were recognized, we selected the top 5 largest blobs in each image and extract the region features in the original image. There were 9 features extracted for each region: average region intensity, stdev of region intensity, region area, region filled area, region perimeter, equivalent circular diam- eter, convex hull area, eccentricity, and regionâs extent [13, 16]. As a result, 52 features were extracted from each image: f eatures= 7 binary features +(9 region features)Ă(5 large regions) (1) 6Truong X. Tran et al. TFL trial Images Data splitter Training Images Model training (include validation) Model Testing Final Model New wet-lab trial images Predicted category Testing Images Fig. 7The framework of classifying TFLâd protein crystal- lization images. 3.5 Building Classification Model Using Image Segmentation Features The framework for building our classification model is provided in Fig. 7. There are three major stages in the framework: i) model training and validation, i) model testing, and i) final model evaluation. These steps are summarized as follows: âFirst, the TFL2756 data is randomly divided into training and testing data sets by stratified sampling. âIn the model training and validation stage, the train- ing images are segmented by a binarization method. Then, the region features are extracted. The classi- fication model is trained on the extracted features set. The model also is evaluated in the training pro- cess, such as using 10-fold cross-validation. âIn the model testing stage, the testing data set is used to evaluate the classification model. The ex- tracted features in each testing image are fed to the classification model and the prediction result is com- pared with the actual class label. âFinally, in our study, we use whole TFL2756 data to train our final classification model. The final model is used to classify the new wet-lab protein crystal- lization images. 4 Proposed Method: Protein Crystallization Segmentation and Classification Using Subordinate Color Channel We develop a new method to analyze the protein crys- tallization trial images using a subordinate color chan- nel rather than the dominant color in the fluorescence microscopy images. We first introduce the concept of subordinate color and the basics of TFL image intensity analysis. Then we explain our subordinate percentile intensity thresholding method for segmenting TFL im- ages in the subordinate color channel. 4.1 Subordinate Color Channel in Fluorescence Emission In fluorescence microscopy, the sample is usually la- beled with a fluorophore that could emit light when illuminated. In order to achieve an image with maxi- mum fluorescence intensity, the emission wavelengths near or at the peak of the emission are selected for detection. The selected wavelengths create the domi- nant color in the image achieved by the fluorescence microscopy. In most cases the higher the intensity, the better the image analysis result. However, the analy- sis of TFLâd protein crystallization trial images in the dominant color channel is not always easy. Although protein crystals generally have the highest intensity in an image, sometimes there might be bright spots that may not be crystals [12]. Moreover, in some images, crystals appear within a large lower intensity area. In some images, crystals might be out of focus and blurry. Since segmentation and classification of those images are challenging, this problem motivated us to analyze other wavelengths than the the emission wavelength. Consider the spectrum of excitation and emission lights for Carboxyrhodamine 6G - Succinimidyl Ester (CR-6G, C6157) illustrated in Fig. 8. The CR-6G has excitation and emission spectra in between those of fluorescein and tetramethylrhodamine and have higher fluorescence quantum yield (the ratio of absorption to emission photons) than tetramethylrhodamine conju- gates. With the peak absorption of CR-6G at 525nm, we use a 530nm laser light in our microscope as the illumination source. As shown in the figure, the peak intensity of the emission when illuminated by a 530nm light is around 550nm wavelength. However, CR-6G also emits light for wavelengths greater than 590nm de- spite having lower intensity in images. If the far right side of Fig. 8 for wavelengths from 595-700nm is con- sidered as a long tail, this side has intensity values in the red channel of RGB images. The relative intensity interval is fromâŒ28% to 2%. Protein Crystallization Segmentation and Classification Using Subordinate Color Channel7 We consider the emitted light corresponding to the longer wavelength (red channel in our case) with low intensity as the subordinate color channel in our fluo- rescence image. In Fig. 8, it can be seen that there is not much emission for wavelengths less than 500nmand do we do not expect any intensity for these wavelengths corresponding to bluish colors. 4.2 Color Intensity Analysis of The Protein Crystallization Trial Images We propose using other color channels based on the filter used on the microscope rather than using the pri- mary color to analyze the protein crystallization trial images. We hypothesize that the analysis of other color channels may provide better results than the analysis of the primary color channel. In the TFL2756 data sets, the green color is dom- inant and the red color is subordinate. Fig. 9 displays the pixel values at crystal and non-crystal regions in RGB trial images of a clear drop sample (unsuccessful trial) and a protein crystal (successful trial). In this ex- ample, green color has a high intensity value (>210) in both crystal and non-crystal images. If non-crystal and crystal images are analyzed, the green intensity varia- tion is not very significant for crystal images and non- crystal images since in another image it is possible to have a crystal region with a lower green intensity than in a non-crystal region in another image. It is desirable that the intensity change is significant between a crys- tal region and non-crystal region across images. If the subordinate red color is considered for comparison, the change in red color intensity (around 34 in the non- crystal image whereas>225 in the crystal image) is highly different and it could serve as a better candidate to distinguish crystal and non-crystal regions. To further analyze the subordinate color channel, for each image (I hĂw ) in non-crystal, likely-lead, and crystal categories, the maximum green and red intensity values are taken as follows: I Green max =M AX 1â€iâ€h,1â€jâ€w (I Green (i, j)),(2) I Red max =M AX 1â€iâ€h,1â€jâ€w (I Red (i, j)).(3) Then for dominant color channel, green, and subordi- nate color channel, red, we analyzed the distribution of maximum pixel intensity for each image category. Fig. 10 and Fig. 11 display the histogram ofI Red max and I Green max grouped by non-crystal, likely-lead, and crystal categories. In the subordinate red channel, almost all non-crystal images have lower maximum intensity than crystals and likely-leads. For the dominant green chan- nel, the highest intensity value of non-crystal images spread all over the intensity scale. This indicates an overlap for the green channel between non-crystal and other categories. These figures show that the subordi- nate red channel could be more suitable for classifying images than the dominant green channel. We further analyzed the overall intensity of images. The histogram of the average gray intensity is plotted in Fig.12. It can be seen that the average gray inten- sity has similar distributions for all categories. Then we have used the threshold of Otsuâs thresholding algo- rithm [9] and plotted the histogram of the thresholds in Fig.13. Although Otsuâs method provides better distri- bution, the right side of the peak for non-crystal images have overlaps with crystal and and likely-lead images. Hence, we may state that the subordinate color chan- nel is more promising than the dominant color channel and the overall intensity while distinguishing image cat- egories. 4.3 Subordinate Color Percentile Intensity Thresholding We have developed a percentile-based thresholding method named subordinate color percentile intensity threshold- ing (SPI) to segment the fluorescence microscopy im- ages using a subordinate color channel. In general, the SPI method could generate binary images using any color channel. However, in our study, SPI creates the bi- nary image using the intensity of the subordinate color in the protein crystallization trial images (the red color in this case). The SPI thresholding has three threshold values. The most important factor is thecolor percentile thresh- old,T CP , that satisfies the percentile intensity condition over all pixels in an image. For instance, if the percentile is chosen as 90% for a channel, the percentile threshold is selected in a way that 90% of intensity values are be- low this threshold. Let the color channelCof an image IbeI C . TheP th percentile intensity threshold of the Ccolor inIis computed using the histogram as: T CP =M ax(k)| k X i=0 histogram(I C , i)< PĂhĂw(4) The second threshold value,t min , is the minimum intensity requirement of a crystal image. There must be some high-intensity regions in an image of a crystal. The crystal regions should have intensity high enough compared to the background. For this purpose, SPI uses t min threshold to remove any low-intensity pixels. This threshold is based on the color-channel intensity instead of the gray-level intensity. A pixeliin the imageIis 8Truong X. Tran et al. Excitation: - - - Emission: Fig. 8Fluorescence spectra (simulated using SpectraViewer tool[7]) when illuminating Carboxyrhodamine 6G - Succinimidyl Ester (CR-6G) with a 530nm Laser light. Non- Crystal Crystal Fig. 9Inspection pixel values for non-crystal and crystal regions. 0 0.05 0.1 Max Red Color Chanel Intensity 0.15 0.2 0 25 0.25 50 75 100 125 150 175 200 225 250 likely-lead non-crystal crystal Fig. 10Histogram of maximum red intensity for all three image categories. assigned to background region if the pixel intensity for colorCis less than or equal tot min threshold: I(i) = ( background,ifI C (i)â€t min , evaluate furtherotherwise. (5) The third threshold is a relaxing thresholding and denoted as t max . Sometimes, the percentile threshold could be a little high and crop parts of crystal regions. To relax the percentile threshold, t max is introduced. Even though a pixelI(i) intensity in theCcolor is less than the percentile threshold, it may be chosen as a 0 0 0.05 25 0.1 0.15 50 0.2 0.25 75 0.3 0.35 100 0.4 0.45 125 0.5 Max Green Color Chanel Intensity 0.55 150 0.6 0.65 175 0.7 0.75 200 225 250 crystal likely-lead non-crystal Fig. 11Histogram of maximum green intensity for all three image categories. foreground pixel if its intensity value is greater than or equal to t max . I(i) = ( foreground,ifI C (i)â„t max , evaluate furtherotherwise. (6) The SPI image binarization equation can be ex- pressed as: I(i) = ïŁ± ïŁŽ ïŁČ ïŁŽ ïŁł f oreground,ifI C (i)> t min andI C (i)> T CP orI C (i)â„t max background,otherwise. (7) Protein Crystallization Segmentation and Classification Using Subordinate Color Channel9 0 0 0.05 25 0.1 50 75 0.15 100 125 Avegrate Gray Image Intensity 150 175 200 225 250 non-crystal crystal likely-lead Fig. 12Histogram of average gray image intensity for all three image categories. 0 Global Image Threshold Intensity 0.05 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.1 crystal likely-lead non-crystal Fig. 13The global intensity histogram threshold using Otsuâs method for all three image categories. The procedure of SPI thresholding is presented in Algorithm 1. The SPI algorithm takes an RGB image Ias an input. The subordinate color channelC, per- centile intensityP,t min , andt max are the formal pa- rameters of the algorithm. SPI output the binary image Bof the original imageI. The following steps summa- rizes the algorithm: âStep 1 (line 1): Obtain theI C intensity matrix for the colorCfrom the original imageI. âStep 2 (line 2): Compute theP th percentile intensity thresholdT CP of the colorCusing formula (4). âStep 3 (line 3 -6): Create the binary image. The pixeliinI C is assigned to foreground region if the colorCintensityI C (i) greater than botht min and T CP or greater than or equal tot max . Otherwise, it is assigned to the background region (7)). We are going to use the following notation to de- note SPI thresholding of a color channelCat a par- ticular percentile intensityP:Subfollowed by the color channel and the color channel followed by the percentile as SubCP. For example,SubR90 means SPI threshold- Algorithm 1SPI Input:Original Image:I Input:Color channel:C Input:Percentile intensity:P Input:Min and max intensity thresholds of colorC: t max , t min Output:Binary image:B 1: Obtain channelI C fromIunder colorC 2: Compute thresholdT CP thatPpercent of pixels inI C has colorCintensity<T CP 3: FOREACH pixeliinI C with intensityI C (i) 4: IF (I C (i)>t min &I C (i)>T CP )k(I C (i)>= t max ) THEN 5:Set pixelias foreground 6: ELSE set pixelias background ing on the subordinate red channel (SubR) with 90 th percentile. In addition,SubR90 uses (t max ) and (t min ) thresholding values. There is one significant difference between the SPI method we develop for subordinate color channel and the dominant green percentile thresholding method de- scribed in Section 3.B. t min of dominant green per- centile thresholding is based on the gray level. When we adapt t max threshold to the original dominant green percentile intensity method, this threshold is applied for gray level. For example, consider a pixeliin dominant green channelI G with intensityI G (i)and general gray levelI gray (i). This pixel is set as a foreground pixel for green percentile intensity thresholding using three threshold values, (t max ), (t min ), and T GP if the follow- ing condition is satisfied: (I gray (i)> t min andI G (i)> T GP )or(I gray (i)â„t max ). 5 Experiments and Results All experiments were performed on CR-6G fluorophore attached to proteins of interest. Our Crystal X2 flu- orescence microscope is used to capture the screening samples. The image segmentation and feature extrac- tion were implemented using MATLAB (2018a), and the classification models are built onTFL276 The experiments in this section examine the clas- sification of protein crystallization trials based on SPI segmentation method. We analyze the performance of SPI method based on the subordinate channel and eval- uate how it performs with respect to the dominant color channel. We evaluate its performance on classification and also provide how it can benefit from the variation of thresholding values. Then, we provide the overall clas- sification performance of using the subordinate color channel for segmentation and feature extraction. Fi- nally, we have performed additional new wet-lab ex- periments and check the performance of using the sub- ordinate color to determine its generalizability. 10Truong X. Tran et al. While evaluating the classification performance and thresholding values, we used 90% of the image data set for training and the remaining 10% of the data for final evaluation of the optimized model. The images were randomly split but the ratio of image categories were preserved as in stratified sampling. 5.1 Evaluation Measures The objective of our research is to minimize the rate of missing a crystal, in other words, the misclassification of any crystal or leading condition as a non-crystal. The evaluation measures that are used in this study are as follows: 1. Misclassification of both likely-leads and crystals as non-crystal rate (lower is better): M LC= ml+mc l+c (8) 2. Misclassification of crystals as non-crystal rate (lower is better): M C= mc c (9) 3. Accuracy (higher is better): ACC= cn+cl+c n+l+c (10) wherenis the number of non-crystals,lis the num- ber of likely-leads,cis the number of crystals,cnis the number of correctly classified non-crystals,clis the number of correctly classified likely leads,ccis the num- ber of correctly classified crystals,mlis the number of likely-leads misclassified as non-crystal, andmcis the number of crystals misclassified as non-crystal. 5.2 Performance of the Subordinate Color Channel on Classification In this section, we compare the performance of our SPI method using the subordinate channel to other thresh- olding methods relying on the dominant color channel or overall intensity. Hence, to experimentally investi- gate the effectiveness of the proposed SPI threshold- ing method, we trained and evaluated the performance of different classifier algorithms on the feature sets ex- tracted from SPI and other thresholding methods. More specifically, we compare the SPI method using the subordinate color channel with dominant green per- centile thresholding and the Otsuâs thresholding using the overall intensity. The Otsuâs thresholding method calculates its optimum threshold value based on the as- sumption of bimodal histogram distribution[9]. Otsuâs thresholding does not require t min or t max parameters. For the dominant green percentile intensity threshold- ing method, we used four percentiles for thresholding as DomG90, DomG97, DomG99, and DomG100. The t min was set equal to 40 as reported in our previous research[4, 5, 13, 14, 16] and t max was set to 255 (the highest value on grayscale). The proposed SPI method is performed on the subordinate red color channel with four percentiles for thresholding of SubR90, SubR97, SubR99, and SubR100. Initially, we set t min = 0, the lowest red intensity value, and set t max equal to 255. The TFL2756 images were segmented by each thresh- olding method. Then, the features were extracted from each image to create the feature set (see Section 3.4). These feature sets were used to train and evaluate the classification performance. Each feature set has 52 at- tributes and a class label of non-crystal, likely-lead, or crystal category. For classification models, we used four algorithms: Naive Bayes (NB), which is a classifier using Bayes the- orem with assumptption that a particular feature is in- dependent of other features; Decision Tree (DT), which we used C4.5 algorithm and the pruning condition was set as (minimum) 2 instances per leaf; Random Forest (RF), which was set to have 100 trees; and Multilayer Perceptron Neural Networks (MLP), which was con- structed with one hidden layer of 27 neurons, the input layer of 52 neurons, and the output of 3 nodes. Among the four classifiers, NB had the lowest ACC (<80%) and a high MLC rate (averaging around 15.30% for the dominant color channel using DomG90, DomG97, DomG99, and DomG100) as shown in Fig. 14 and Fig. 16. Using the NB classifier, SubR90 achieved MLC rate close to 0 and accuracy close to 85%. Among all clas- sifiers RF performed well for both the dominant and subordinate color channels. However, the MLC rate of SubR90 was 0.96 (below 1) whereas the MLC rate of DomG99 was only 3.75. Using the subordinate chan- nel not only outperformed for RF and NB, but it also performed better for MLP and DT classifiers. These re- sults show that the superiority of using the subordinate channel is across four classifiers used in this study. Compared to using the dominant color channel, our segmentation method on subordinate color channel was able to the reduce the misclassification rate of likely- leads or crystals as non-crystals by the percentage of from 9.71% to 2.02% depending on the classifier. Sim- ilarly, the accuracy of classifiers were increased by the percentage of from 1.77% to 5.53%. Our method reached 93.91% accuracy while keeping misclassification of likely- leads and crystals as non-crystals below 1%. When us- ing our method, the rate of a crystal misclassified as non-crystal (MC) is also decreased compared to the Protein Crystallization Segmentation and Classification Using Subordinate Color Channel11 O#"$ $ $ $ $ # ""! #"$ $ $ $ $ # ""! #"$ $ $ $ $ # ""! #"$ $ $ $ $ # ""! Fig. 14The rate of misclassifying crystal and likely-lead as non-crystal (MLC) of NB, DT, RF, and MLP. O! " """ " ! ! " """ " ! ! " """ " ! ! " """ " ! Fig. 15The rate of misclassifying crystal as non-crystal (MC) of NB, DT, RF, and MLP. O ! !!! ! !" ! !!! ! !" ! !!! ! !" ! !!! ! !" Fig. 16The accuracy of NB, DT, RF, and MLP. method utilizing the dominant color channel (see Fig. 15). Among the SPI features sets, SubR99 stands out as the best data set yielding high accuracies and low MLC rates. SubR99 results MLC rate of around 1% with an accuracy ofâŒ93.5% for RF andâŒ92.0% for MLP classifiers. 12Truong X. Tran et al. 0 95 0.01 0.02 0.03 MLC rate 0.04 0.05 127 79 63 159 t max t min 47 191 31 15 223 0 255 Fig. 17MLC rate of RF classifier for Dom99 by varying thresholds. 0 0.01 0.02 95 0.03 0.04 0.05 0.06 MLC rate 0.07 0.08 0.09 0.1 127 79 63 159 t max t min 47 191 31 15 223 0 255 Fig. 18MLC rate of RF classifier for SubR99 by varying thresholds. 5.3 Tunability of Thresholding Factors for the Subordinate Channel The goal of this experiment to determine if tuning the threshold values benefit the classification performance and shows possible improvements while binarizing the images. To investigate the impact of threshold factors we analyze 36 pairs oft min andt max parameters on the proposed SPI algorithm. We also evaluate the dominant green percentile intensity thresholding for comparison. Based on the results of the previous experiments, we examine SubR99 and DomG99 in this section. There are 6 different values of intensity for each threshold factor with an interval of 32 fort max from 95 to 255 and 16 fort min from 0 to 79. To evaluate the threshold values, we trained two classification models, RF and MLP, on feature sets extracted for each thresh- olding iteration. The models were evaluated using 10- fold cross-validation. Totally, 72 models were trained and tested. Fig. 17 displays the MLC rate of the dominant color channel (DomG99) using the RF classifier in 3D bar chart with respect to different values of t max and t min . Fig. 19 shows the similar chart for the MLP classifier. In both charts, the MLC rates of DomG99 for RF and 0 95 0.01 0.02 127 0.03 MLC rate 79 0.04 0.05 63 159 t max 47 t min 191 31 15 223 0 255 Fig. 19MLC rate of MLP classifier for DomG99 by varying thresholds. 0 95 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 MLC rate 0.14 0.15 0.16 0.17 0.18 0.19 0.2 127 79 63 159 t max t min 47 191 31 15 223 0 255 Fig. 20MLC rate of MLP classifier for SubR99 by varying thresholds. MLP are between 0.45 and 0.49. It can be observed that DomG99 did not benefit much from the variation oft min andt max parameters. The MLC rates of the subordinate channel using RF and MLP classifiers with respect to the SPI threshold- ing (SubR99) are plotted in Fig. 18 and 20. The 3D bar charts show the effects oft min andt max factors in SubR99 on reduction of the MLC rate. The MLC rate of RF is reduced from a high rate of above 9% to a low rate of below 1%. The lowest MLC rate for RF classifier is just right above 0.5% when t max = 159 and t min = 0. For the MLP classifier, the MLC rate is reduced from above 17% to around 1%. These results show that the dominant color chan- nel could not benefit from varying thresholding values, which imply that methods relying on global thresholds may not be improved much if the dominant color chan- nel is used. On the other hand, the thresholding values had a significant positive impact on the subordinate color channel and showed the room for possible im- provements while binarizing the images and extracting features. Protein Crystallization Segmentation and Classification Using Subordinate Color Channel13 Table 1Classification model testing result Hold-out 90% 3-fold 5-fold 10-fold ACC (%)94.5693.75 93.6594.30 MLC (%)00.570.430.43 MC (%)00.620.620.62 *Data table used with permission from Tran et al. [19]), ©2017 IEEE. 5.4 Overall Performance using the Subordinate Color Channel In this section, we experimentally tested the perfor- mance of using the subordinate color channel by pick- ing the best classifier and best set of threshold values based on the experiment results in the previous sec- tions. Hence, for subordinate color channel using SPI method based on SubR99, we picked RF classifier. This classifier is trained on the feature sets extracted by SubR99 witht max = 159 andt min set to 0. Table 1 shows the performance of SubR99 on TFL2756 data sets using 90% hold-out, 3-fold, 5-fold, and 10-fold cross-validation. The average accuracy of all tests was around 94%. When holding-out 10% of TFL2756 data for testing and training on 90%, the MLC and MC rates of the classification model are both 0. In this case, our model did not miss any crystal or likely-leads condition in the test images. The MLC rates were 0.57%, 0.43%, and 0.43% for 3-fold, 5-fold, and 10-fold cross valida- tion. All the cross-validation tests have the same MC rate of 0.62%. It can be observed that our method is stable and robust on the TFL2756 protein crystalliza- tion trial images. 5.5 Evaluation of the Subordinate Channel on New Wet-lab Experiments To evaluate the generalizability of using the subordi- nate color channel, we have conducted new wet-lab ex- periments and collected new trial images on two pro- teins, Concanavalin A (ConA) and Ab inorganic py- rophosphatase (AbIPPase). Each protein was trace flu- orescently labeled using Carboxyrhodamine 6G. Fig. 21 shows the sample of the screening plates used in this ex- periment. For ConA protein, 96-well screen plate was used (8 rows by 12 columns of single sample well) giv- ing 96 trial images. For AbIPPase, 288-well plate was used giving 288 trial images per scanning. 288-well plate has 3 sub-wells for each 96 well location (96*3=288). We used the sitting-drop vapor-diffusion for the crys- tallization. The trial plates were set up using standard procedures (see Section 3.1). We used the Crystal X2 automated fluorescence mi- croscopy to capture the images of the wells under the il- Fig. 21Sample image of the 96-well plate (top) and 288-well plate (bottom). lumination light at 530nm. These two experiments gen- erated 384 (96+288) trial images. The expert manually reviewed the trial images and assigned scores to each image. For wet-lab experiments, we have achieved an accuracy of 90.36%, MLC of 1.42%, and MC of 0.60%. These results show that the classifier built using the subordinate color channel is generalizable. 6 Discussion Our experiment results using the subordinate color chan- nel show that our method yields high accuracy and low misclassification rate. These results confirm our obser- vation of the preliminary analysis that the use of sub- ordinate color channel provides better result than the primary dominant channel. In trace fluorescent label- ing, the fluorescence dyes will most likely attach to pro- tein crystals. Since the fluorescence dye emits a wide range of light wavelengths, as shown in Fig. 8, the pro- tein crystal could emit at different wavelengths other than the peak emission wavelength. Other chemicals such as salt crystals or precipitates do not attach to the dye molecules. However, because of free floating dye molecules or other reflective chemicals, it is possible to observe the reflection for non-crystal protein regions as well. Therefore, analyzing the fluorescence image be- 14Truong X. Tran et al. Fig. 22Binary images of crystals (a,e) conducted by differ- ent thresholdings: Otsu (b, f); Green channel (c: DomG99, g: DomG97 with t max = 255 and t min = 40)); Red channel ( d: SubR99, h: Sub97 with t max = 255 and t min = 0). yond color channel corresponding to the illumination color may increase the robustness of the analysis. 6.1 Comparing Binarized Images of Dominant and Subordinate Color Channels Thresholding methods can generate proper binarization if images have clearly separated regions as in Fig. 6 (h). The binarization results of Otsu, DomG99 and SubR99 thresholding methods are shown in Fig. 22. Any of the thresholding methods provided good segmentation for this image. Crystal or foreground regions are not always eas- ily separable from the background for images shown in the first row of Fig. 23. Otsuâs thresholding method generates a large foreground area that includes back- ground or non-crystal regions as shown in Fig. 23(d-f). While DomG90 makes a slight improvement with re- spect to Otsuâs method by cropping some of the back- ground or non-crystal regions as shown in Fig. 23(g-i), its output images would be deemed incorrect. The other three methods, DomG99, SubR90, and SubR99, have produced better results. DomG99 still generates fore- ground regions that belong to the background (Fig. 23(j)); however, its crystal binarization shown in Fig. 23(l) would be acceptable. The segmentation images of SubR90 are shown in Fig. 23(n-o). The best results are gen- erated by SubR99 as shown in Fig. 23(p-r) which re- moved non-crystal areas and background quite well as in Fig. 23(p) and had a clear crystal region as depicted in Fig. 23(r). It is visually observed that using the sub- ordinate color channel generates better segmentation results. 6.2 Misclassified Trial Images Even though the experiment results show that our method has high accuracy and low misclassification rate, there are still leading crystallization conditions that are mis- classified as non-crystals. Fig. 24 displays some images Fig. 23Binary images conducted by different thresholding: (a) original non-crystal; (b) original likely lead; (c) origi- nal crystal; (d - f) Otsu thresholding; (g-i) G90, and (j-l) G99 where t max = 255, t min = 40; (n-o) SubR90, and (p- r) SubR99 where t max = 255, t min = 0 (Image used with permission from Tran et al. [19],©2017 IEEE). of likely-lead and crystal that were misclassified as non- crystal images. These misclassification outcomes due to low intensity images or large bright background re- gions. Using proper lighting and in-focus images could improve the image quality, and methods such as Fo- cusALL [17] could be used to generate in-focus images from a stack of images. 6.3 Building Multiple Models for Varying Intensities. Our model is built based on a training set that includes images that have varying intensities. Instead of building one model (i.e., one threshold parameter set), several models could be built for images whose average intensi- ties are within a range. Whenever an input trial image is provided, the learning model for the corresponding in- Protein Crystallization Segmentation and Classification Using Subordinate Color Channel15 Fig. 24Sample of trial images were misclassified as non- crystal: (a-b) low intensity needle crystal; (c) birefringent pre- cipitate likely-lead (Image used with permission from Tranet al. [19],©2017 IEEE). tensity can be used to improve performance. Normally, this would not be needed; however, there could be light- ing problems during image acquisition and captured im- ages may have different intensities than usual. 6.4 Evaluation of Thresholds There are three thresholds used in our method: per- centile threshold,t min andt max . Fig. 18 and Fig. 20 show that MLC rate reduces ast min reduces. For the samet min value, differentt max values yield the same MLC rate. It may be observed that we may chooset min close to 0, and pick at max value. Our method may only uset min and the percentile threshold,T CP and can get the same performance. 7 Conclusion In this paper, we introduced the use of a subordinate channel besides the color channel corresponding to the dominant wavelength in the emission spectra of fluores- cence dye for classifying protein crystallization trial im- ages. We have developed a segmentation method based on the intensity of the subordinate color channel us- ing three thresholds. By using the subordinate channel, we were able to reduce the misclassification of crystals below 1%. Our results indicate that using the subor- dinate channel in the emission spectra could be more helpful in distinguishing crystals than the color chan- nel corresponding to the dominant wavelength. Using the subordinate channel had higher accuracy and lower MLC rates when four classifier models were used. More- over, the segmentation based on the subordinate color channel also benefited from the variation of threshold values. Acknowledgment This research was supported by National Institutes of Health (GM116283) grant. This paper is an extension of our previous work âT. X. Tran and R. S. Aygun and M. L. Pusey, Classify- ing protein crystallization trial images using subordi- nate color channel, in 2017 IEEE Int. Conf. on Bioin- formatics and Biomedicine (BIBM), Nov. 2017, page 1546-1553, DOI 10.1109/BIBM.2017.8217890â.©2017 IEEE. Reprinted, with permission from T. X. Tran and R. S. Aygun and M. L. Pusey, Classifying protein crys- tallization trial images using subordinate color chan- nel, in 2017 IEEE Int. Conf. on Bioinformatics and Biomedicine (BIBM), Nov. 2017, page 1546-1553, DOI 10.1109/BIBM.2017.8217890. References 1. Bern M, Goldberg D, Stevens RC, Kuhn P (2004) Automatic classification of protein crystallization images using a curve-tracking algorithm. Journal of Applied Crystallography 37(2):279â287, DOI 10.1107/S0021889804001761 2. Bruno AE, Charbonneau P, Newman J, Snell EH, So DR, Vanhoucke V, Watkins CJ, Williams S, Wil- son J (2018) Classification of crystallization out- comes using deep convolutional neural networks. PLOS one 13(6):e0198883 3. Cumbaa CA, Jurisica I (2010) Protein crystalliza- tion analysis on the world community grid. Journal of Structural and Functional Genomics 11(1):61â 69, DOI 10.1007/s10969-009-9076-9 4. Dinc I, Dinc S, Sigdel M, Sigdel M, Pusey M, Ay- gun R (2014) Dt-binarize: A hybrid binarization method using decision tree for protein crystalliza- tion images. In: Proceedings of the 2014 Int. Conf. on Image Processing, Computer Vision, Pattern Recognition, p 304â311 5. Dinc I, Dinc S, Sigdel M, Sigdel MS, Pusey ML, Ayg Ìun RS (2017) Super-thresholding: Su- pervised thresholding of protein crystal images. IEEE/ACM Transactions on Computational Bi- ology and Bioinformatics 14(4):986â998, DOI 10.1109/TCBB.2016.2542811 6. Forsythe E, Achari A, Pusey ML (2006) Trace flu- orescent labeling for high-throughput crystallogra- phy. Acta Crystallographica Section D 62(3):339â 346, DOI 10.1107/S0907444906000813 7. Inc TFS (2018) Fluorescence spectraviewer. URL https://w.thermofisher.com/us/en/home.html 8. Klijn ME, Hubbuch J (2019) Time-dependent multi-light-source image classification combined with automated multidimensional protein phase diagram construction for protein phase behavior analysis. Journal of pharmaceutical sciences 9. Otsu N (1979) A threshold selection method from gray-level histograms. IEEE Transactions on Sys- tems, Man, and Cybernetics 9(1):62â66, DOI 10.1109/TSMC.1979.4310076 16Truong X. Tran et al. 10. Pan S, Shavit G, Penas-Centeno M, Xu DH, Shapiro L, Ladner R, Riskin E, Hol W, Meldrum D (2006) Automated classification of protein crystal- lization images using support vector machines with scale-invariant texture and Gabor features. Acta Crystallographica Section D 62(3):271â279, DOI 10.1107/S0907444905041648 11. Po MJ, Laine AF (2008) Leveraging genetic al- gorithm and neural network in automated pro- tein crystal recognition. In: 2008 30th An- nual Int. Conf. of the IEEE Engineering in Medicine and Biology Society, p 1926â1929, DOI 10.1109/IEMBS.2008.4649564 12. Pusey M, Barcena J, Morris M, Singhal A, Yuan Q, Ng J (2015) Trace fluorescent labeling for protein crystallization. Acta Crystallographica Section F 71(7):806â814, DOI 10.1107/S2053230X15008626 13. Sigdel M, Pusey ML, Aygun RS (2013) Real- time protein crystallization image acquisition and classification system. Crystal Growth & Design 13(7):2728â2736, DOI 10.1021/cg3016029 14. Sigdel M, Sigdel M, Dinc I, Dinc S, Pusey M, Ay- gun R (2014) Classification of protein crystalliza- tion trial images using geometric features. In: Pro- ceedings of the 2014 Int. Conf. on Image Processing, Computer Vision, Pattern Recognition, p 192â198 15. Sigdel M, Pusey ML, Aygun RS (2015) Crystpro: Spatiotemporal analysis of protein crystallization images. Crystal growth & design 15(11):5254â5262 16. Sigdel M, Dinc I, Sigdel MS, Dinc S, Pusey ML, Aygun RS (2017) Feature analysis for classifica- tion of trace fluorescent labeled protein crystal- lization images. BioData Mining 10(1):14, DOI 10.1186/s13040-017-0133-9 17. Sigdel MS, Sigdel M, Din ̧c S, Dinc I, Pusey ML, Ayg Ìun RS (2016) Focusall: Focal stacking of micro- scopic images using modified harris corner response measure. IEEE/ACM Transactions on Computa- tional Biology and Bioinformatics 13(2):326â340, DOI 10.1109/TCBB.2015.2459685 18. Spraggon G, Lesley SA, Kreusch A, Pries- tle JP (2002) Computational analysis of crystallization trials. Acta Crystallograph- ica Section D 58(11):1915â1923, DOI 10.1107/S0907444902016840 19. Tran TX, Aygun RS, Pusey ML (2017) Classify- ing protein crystallization trial images using subor- dinate color channel. In: 2017 IEEE Int. Conf. on Bioinformatics and Biomedicine (BIBM), vol 00, p 1546â1553, DOI 10.1109/BIBM.2017.8217890 20. Tran TX, Pusey ML, Aygun RS (2018) Else-tree classifier for minimizing misclassification of biolog- ical data. In: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), IEEE, p 2301â2308 21. Zhu X, Sun S, Bern M (2004) Classification of pro- tein crystallization imagery. In: The 26th Annual Int. Conf. of the IEEE Engineering in Medicine and Biology Society, vol 1, p 1628â1631, DOI 10.1109/IEMBS.2004.1403493 Author Biography Truong X. Tranreceived his Bach- elorâs degree major in Electronic and Telecommunication Engineering, minor in Biomedical Electronic Engineering, from Hanoi University of Science and Technology, Vietnam in 2010, the M.S. degree in Computer Science from Arkansas State University, AR, USA in 2012. He is pursuing Doctoral degree in Computer Sci- ence at the University of Alabama in Huntsville, AL, USA. He works in Data analysis for protein crystal- lization research project. His research interests include machine learning and data mining. Marc L. Puseyis a research sci- entist working at iXpressGenes, Inc., Huntsville Alabama. Dr. Pusey received his Ph.D. in Biochemistry from the Uni- versity of Miami, Fl., then did post- doctoral research at the University of Minnesota. He moved to Huntsville in 1985, and worked at NASA/MSFC for the next 23 years in the new field of protein crystal growth. After retirement from NASA, he obtained his current position at iX- pressGenes with the focus of his research being im- proved methods for protein crystal screening, crystal detection, and crystallization condition identification using visible fluorescence methods. Ramazan S Ayg Ìun:received the B.S. degree in computer engineering from Bilkent University, Ankara, Turkey in 1996, the M.S. degree from Mid- dle East Technical University, Ankara in 1998, and the Ph.D. degree in com- puter science and engineering from State University of New York at Buffalo in 2003. He is currently an Associate Pro- fessor in Computer Science Department, University of Alabama in Huntsville. His research interests include Protein Crystallization Segmentation and Classification Using Subordinate Color Channel17 protein crystallization image analysis, data mining, im- age and video processing, spatio-temporal indexing and querying, multimedia databases, semantic computing, multimedia networking, and multimedia synchroniza- tion.