Interindividual distances and orientations of laying hens under 8 stocking densities measured by integrative deep learning techniques.

Interindividual distances and orientations of laying hens provide quantitative measures to calculate and optimize space allocations for bird flocks. However, these metrics were often measured manually and have not been examined for different stocking densities of laying hens. The objectives of this study were to 1) integrate and develop several deep learning techniques to detect interindividual distances and orientations of laying hens; and 2) examine the 2 metrics under 8 stocking densities via the developed techniques. Laying hens (Jingfen breed, a popular hen breed in China) at 35 wk of age were raised in experimental compartments at 8 different stocking densities of 3,840, 2,880, 2,304, 1,920, 1,646, 1,440, 1,280, and 1,152 cm2•bird-1 (3-10 hens per compartment, respectively), and cameras on the top of the compartments recorded videos for further analysis. The designed deep learning image classifier achieved over 99% accuracy to classify bird's perching status and excluded frames with bird perching to ensure that all birds analyzed were on the same horizontal plane, reducing calculation errors. The YOLOv5m oriented object detection model achieved over 90% precision, recall, and F1 score in detecting birds in compartments and can output bird centroid coordinates and angles, from which interindividual distances and orientations were calculated based on pairs of birds. Laying hens maintained smaller minimum interindividual distances in higher stocking densities. They were in an intersecting relationship with conspecifics for over 90% of the time. The developed integrative deep learning techniques and behavior metrics provide animal-based measurement of space requirement for laying hens.

Supervised and Weakly Supervised Deep Learning for Segmentation and Counting of Cotton Bolls Using Proximal Imagery.

The total boll count from a plant is one of the most important phenotypic traits for cotton breeding and is also an important factor for growers to estimate the final yield. With the recent advances in deep learning, many supervised learning approaches have been implemented to perform phenotypic trait measurement from images for various crops, but few studies have been conducted to count cotton bolls from field images. Supervised learning models require a vast number of annotated images for training, which has become a bottleneck for machine learning model development. The goal of this study is to develop both fully supervised and weakly supervised deep learning models to segment and count cotton bolls from proximal imagery. A total of 290 RGB images of cotton plants from both potted (indoor and outdoor) and in-field settings were taken by consumer-grade cameras and the raw images were divided into 4350 image tiles for further model training and testing. Two supervised models (Mask R-CNN and S-Count) and two weakly supervised approaches (WS-Count and CountSeg) were compared in terms of boll count accuracy and annotation costs. The results revealed that the weakly supervised counting approaches performed well with RMSE values of 1.826 and 1.284 for WS-Count and CountSeg, respectively, whereas the fully supervised models achieve RMSE values of 1.181 and 1.175 for S-Count and Mask R-CNN, respectively, when the number of bolls in an image patch is less than 10. In terms of data annotation costs, the weakly supervised approaches were at least 10 times more cost efficient than the supervised approach for boll counting. In the future, the deep learning models developed in this study can be extended to other plant organs, such as main stalks, nodes, and primary and secondary branches. Both the supervised and weakly supervised deep learning models for boll counting with low-cost RGB images can be used by cotton breeders, physiologists, and growers alike to improve crop breeding and yield estimation.

Quantitative Structure-Mutation-Activity Relationship Tests (QSMART) model for protein kinase inhibitor response prediction.

BACKGROUND: Protein kinases are a large family of druggable proteins that are genomically and proteomically altered in many human cancers. Kinase-targeted drugs are emerging as promising avenues for personalized medicine because of the differential response shown by altered kinases to drug treatment in patients and cell-based assays. However, an incomplete understanding of the relationships connecting genome, proteome and drug sensitivity profiles present a major bottleneck in targeting kinases for personalized medicine. RESULTS: In this study, we propose a multi-component Quantitative Structure-Mutation-Activity Relationship Tests (QSMART) model and neural networks framework for providing explainable models of protein kinase inhibition and drug response ([Formula: see text]) profiles in cell lines. Using non-small cell lung cancer as a case study, we show that interaction terms that capture associations between drugs, pathways, and mutant kinases quantitatively contribute to the response of two EGFR inhibitors (afatinib and lapatinib). In particular, protein-protein interactions associated with the JNK apoptotic pathway, associations between lung development and axon extension, and interaction terms connecting drug substructures and the volume/charge of mutant residues at specific structural locations contribute significantly to the observed [Formula: see text] values in cell-based assays. CONCLUSIONS: By integrating multi-omics data in the QSMART model, we not only predict drug responses in cancer cell lines with high accuracy but also identify features and explainable interaction terms contributing to the accuracy. Although we have tested our multi-component explainable framework on protein kinase inhibitors, it can be extended across the proteome to investigate the complex relationships connecting genotypes and drug sensitivity profiles.

Deep evolutionary analysis reveals the design principles of fold A glycosyltransferases.

Glycosyltransferases (GTs) are prevalent across the tree of life and regulate nearly all aspects of cellular functions. The evolutionary basis for their complex and diverse modes of catalytic functions remain enigmatic. Here, based on deep mining of over half million GT-A fold sequences, we define a minimal core component shared among functionally diverse enzymes. We find that variations in the common core and emergence of hypervariable loops extending from the core contributed to GT-A diversity. We provide a phylogenetic framework relating diverse GT-A fold families for the first time and show that inverting and retaining mechanisms emerged multiple times independently during evolution. Using evolutionary information encoded in primary sequences, we trained a machine learning classifier to predict donor specificity with nearly 90% accuracy and deployed it for the annotation of understudied GTs. Our studies provide an evolutionary framework for investigating complex relationships connecting GT-A fold sequence, structure, function and regulation.

Carbohydrates are one of the major groups of large biological molecules that regulate nearly all aspects of life. Yet, unlike DNA or proteins, carbohydrates are made without a template to follow. Instead, these molecules are built from a set of sugar-based building blocks by the intricate activities of a large and diverse family of enzymes known as glycosyltransferases. An incomplete understanding of how glycosyltransferases recognize and build diverse carbohydrates presents a major bottleneck in developing therapeutic strategies for diseases associated with abnormalities in these enzymes. It also limits efforts to engineer these enzymes for biotechnology applications and biofuel production. Taujale et al. have now used evolutionary approaches to map the evolution of a major subset of glycosyltransferases from species across the tree of life to understand how these enzymes evolved such precise mechanisms to build diverse carbohydrates. First, a minimal structural unit was defined based on being shared among a group of over half a million unique glycosyltransferase enzymes with different activities. Further analysis then showed that the diverse activities of these enzymes evolved through the accumulation of mutations within this structural unit, as well as in much more variable regions in the enzyme that extend from the minimal unit. Taujale et al. then built an extended family tree for this collection of glycosyltransferases and details of the evolutionary relationships between the enzymes helped them to create a machine learning framework that could predict which sugar-containing molecules were the raw materials for a given glycosyltransferase. This framework could make predictions with nearly 90% accuracy based only on information that can be deciphered from the gene for that enzyme. These findings will provide scientists with new hypotheses for investigating the complex relationships connecting the genetic information about glycosyltransferases with their structures and activities. Further refinement of the machine learning framework may eventually enable the design of enzymes with properties that are desirable for applications in biotechnology.

Accurate prediction of human miRNA targets via graph modeling of the miRNA-target duplex.

miRNAs are involved in many critical cellular activities through binding to their mRNA targets, e.g. in cell proliferation, differentiation, death, growth control, and developmental timing. Accurate prediction of miRNA targets can assist efficient experimental investigations on the functional roles of miRNAs. Their prediction, however, remains a challengeable task due to the lack of experimental data about the tertiary structure of miRNA-target binding duplexes. In particular, correlations of nucleotides in the binding duplexes may not be limited to the canonical Watson Crick base pairs (BPs) as they have been perceived; methods based on secondary structure prediction (typically minimum free energy (MFE)) have only had mix success. In this work, we characterized miRNA binding duplexes with a graph model to capture the correlations between pairs of nucleotides of an miRNA and its target sequences. We developed machine learning algorithms to train the graph model to predict the target sites of miRNAs. In particular, because imbalance between positive and negative samples can significantly deteriorate the performance of machine learning methods, we designed a novel method to re-sample available dataset to produce more informative data learning process. We evaluated our model and miRNA target prediction method on human miRNAs and target data obtained from mirTarBase, a database of experimentally verified miRNA-target interactions. The performance of our method in target prediction achieved a sensitivity of 86% with a false positive rate below 13%. In comparison with the state-of-the-art methods miRanda and RNAhybrid on the test data, our method outperforms both of them by a significant margin. The source codes, test sets and model files all are available at http://rna-informatics.uga.edu/?f=software&p=GraB-miTarget .

Classifying kinase conformations using a machine learning approach.

BACKGROUND: Signaling proteins such as protein kinases adopt a diverse array of conformations to respond to regulatory signals in signaling pathways. Perhaps the most fundamental conformational change of a kinase is the transition between active and inactive states, and defining the conformational features associated with kinase activation is critical for selectively targeting abnormally regulated kinases in diseases. While manual examination of crystal structures have led to the identification of key structural features associated with kinase activation, the large number of kinase crystal structures (~3,500) and extensive conformational diversity displayed by the protein kinase superfamily poses unique challenges in fully defining the conformational features associated with kinase activation. Although some computational approaches have been proposed, they are typically based on a small subset of crystal structures using measurements biased towards the active site geometry. RESULTS: We utilize an unbiased informatics based machine learning approach to classify all eukaryotic protein kinase conformations deposited in the PDB. We show that the orientation of the activation segment, measured by φ, ψ, χ1, and pseudo-dihedral angles more accurately classify kinase crystal conformations than existing methods. We show that the formation of the K-E salt bridge is statistically dependent upon the activation segment orientation and identify evolutionary differences between the activation segment conformation of tyrosine and serine/threonine kinases. We provide evidence that our method can identify conformational changes associated with the binding of allosteric regulatory proteins, and show that the greatest variation in inactive structures comes from kinase group and family specific side chain orientations. CONCLUSION: We have provided the first comprehensive machine learning based classification of protein kinase active/inactive conformations, taking into account more structures and measurements than any previous classification effort. Further, our unbiased classification of inactive structures reveals residues associated with kinase functional specificity. To enable classification of new crystal structures, we have made our classifier publicly accessible through a stand-alone program housed at https://github.com/esbg/kinconform [DOI: 10.5281/zenodo.249090 ].

Inhibition of SA Node at Supine Position in Right Atrial Thrombus Complicating Behçet's Disease - From Cardiac Surgical Point of View.

BACKGROUND: Behçet's disease (BD) is a chronic multi-systemic disease of unknown cause. Intra-cardiac thrombus (ICT) complicating BD is extremely rare. In general, cardiac manifestations in BD are associated with poor prognosis. Chest computed tomography (CT) scan and echocardiogram are excellent modalities for diagnosis and patient assessment. Cardiac surgical intervention can be done safely using an on-pump technique when medical management has failed. CASE REPORT: We report on a case of a 27-year-old Jordanian woman diagnosed with BD who presented with dyspnea, cough, and hemoptysis, with supine bradycardia reaching 36 beats/minute and dizziness which disappear on sitting or standing position, and with heart rate reaching 76 beats/minute. Right atrial thrombus was identified using transthoracic echocardiogram and chest CT scan. After medical management failed, cardiac surgical intervention became an option and targeted extraction of the right atrial thrombus compressing the sinoatrial node (SA node). CONCLUSIONS: In BD, right atrial thrombus compressing the SA node is rare. If conservative management has failed, cardiac surgical intervention removing the mechanical cause can be done safely, either using on-pump with cross clamp or on-pump with beating heart technique.

Prediction and prioritization of rare oncogenic mutations in the cancer Kinome using novel features and multiple classifiers.

Cancer is a genetic disease that develops through a series of somatic mutations, a subset of which drive cancer progression. Although cancer genome sequencing studies are beginning to reveal the mutational patterns of genes in various cancers, identifying the small subset of "causative" mutations from the large subset of "non-causative" mutations, which accumulate as a consequence of the disease, is a challenge. In this article, we present an effective machine learning approach for identifying cancer-associated mutations in human protein kinases, a class of signaling proteins known to be frequently mutated in human cancers. We evaluate the performance of 11 well known supervised learners and show that a multiple-classifier approach, which combines the performances of individual learners, significantly improves the classification of known cancer-associated mutations. We introduce several novel features related specifically to structural and functional characteristics of protein kinases and find that the level of conservation of the mutated residue at specific evolutionary depths is an important predictor of oncogenic effect. We consolidate the novel features and the multiple-classifier approach to prioritize and experimentally test a set of rare unconfirmed mutations in the epidermal growth factor receptor tyrosine kinase (EGFR). Our studies identify T725M and L861R as rare cancer-associated mutations inasmuch as these mutations increase EGFR activity in the absence of the activating EGF ligand in cell-based assays.

Decision tree and ensemble learning algorithms with their applications in bioinformatics.

Machine learning approaches have wide applications in bioinformatics, and decision tree is one of the successful approaches applied in this field. In this chapter, we briefly review decision tree and related ensemble algorithms and show the successful applications of such approaches on solving biological problems. We hope that by learning the algorithms of decision trees and ensemble classifiers, biologists can get the basic ideas of how machine learning algorithms work. On the other hand, by being exposed to the applications of decision trees and ensemble algorithms in bioinformatics, computer scientists can get better ideas of which bioinformatics topics they may work on in their future research directions. We aim to provide a platform to bridge the gap between biologists and computer scientists.

Classification of genomic islands using decision trees and their ensemble algorithms.

BACKGROUND: Genomic islands (GIs) are clusters of alien genes in some bacterial genomes, but not be seen in the genomes of other strains within the same genus. The detection of GIs is extremely important to the medical and environmental communities. Despite the discovery of the GI associated features, accurate detection of GIs is still far from satisfactory. RESULTS: In this paper, we combined multiple GI-associated features, and applied and compared various machine learning approaches to evaluate the classification accuracy of GIs datasets on three genera: Salmonella, Staphylococcus, Streptococcus, and their mixed dataset of all three genera. The experimental results have shown that, in general, the decision tree approach outperformed better than other machine learning methods according to five performance evaluation metrics. Using J48 decision trees as base classifiers, we further applied four ensemble algorithms, including adaBoost, bagging, multiboost and random forest, on the same datasets. We found that, overall, these ensemble classifiers could improve classification accuracy. CONCLUSIONS: We conclude that decision trees based ensemble algorithms could accurately classify GIs and non-GIs, and recommend the use of these methods for the future GI data analysis. The software package for detecting GIs can be accessed at http://www.esu.edu/cpsc/che_lab/software/GIDetector/.

Using surrogate modeling in the prediction of fibrinogen adsorption onto polymer surfaces.

We present a Surrogate (semiempirical) Model for prediction of protein adsorption onto the surfaces of biodegradable polymers that have been designed for tissue engineering applications. The protein used in these studies, fibrinogen, is known to play a key role in blood clotting. Therefore, fibrinogen adsorption dictates the performance of implants exposed to blood. The Surrogate Model combines molecular modeling, machine learning and an Artificial Neural Network. This novel approach includes an accounting for experimental error using a Monte Carlo analysis. Briefly, measurements of human fibrinogen adsorption were obtained for 45 polymers. A total of 106 molecular descriptors were generated for each polymer. Of these, 102 descriptors were computed using the Molecular Operating Environment (MOE) software based upon the polymer chemical structures, two represented different monomer types, and two were measured experimentally. The Surrogate Model was developed in two stages. In the first stage, the three descriptors with the highest correlation to adsorption were determined by calculating the information gain of each descriptor. Here a Monte Carlo approach enabled a direct assessment of the effect of the experimental uncertainty on the results. The three highest-ranking descriptors, defined as those with the highest information gain for the sample set, were then selected as the input variables for the second stage, an Artificial Neural Network (ANN) to predict fibrinogen adsorption. The ANN was trained using one-half of the experimental data set (the training set) selected at random. The effect of experimental error on predictive capability was again explored using a Monte Carlo analysis. The accuracy of the ANN was assessed by comparison of the predicted values for fibrinogen adsorption with the experimental data for the remaining polymers (the validation set). The mean value of the Pearson correlation coefficient for the validation data sets was 0.54 +/- 0.12. The average root-mean-square (relative) error in prediction for the validation data sets is 38%. This is an order of magnitude less than the range of experimental values (i.e., 366%) and compares favorably with the average percent relative standard deviation of the experimental measurements (i.e., 17.9%). The effects of each of the user-defined parameters in the ANN were explored. None were observed to have a significant effect on the results. Thus, the Surrogate Model can be used to accurately and unambiguously identify polymers whose fibrinogen absorption is at the limits of the range (i.e., low or high) which is an essential requirement for assessing polymers for regenerative tissue applications.