Geometric potentials from deep learning improve prediction of CDR H3 loop structures

Abstract Motivation Antibody structure is largely conserved, except for a complementarity-determining region featuring six variable loops. Five of these loops adopt canonical folds which can typically be predicted with existing methods, while the remaining loop (CDR H3) remains a challenge due to its highly diverse set of observed conformations. In recent years, deep neural networks have proven to be effective at capturing the complex patterns of protein structure. This work proposes DeepH3, a deep residual neural network that learns to predict inter-residue distances and orientations from antibody heavy and light chain sequence. The output of DeepH3 is a set of probability distributions over distances and orientation angles between pairs of residues. These distributions are converted to geometric potentials and used to discriminate between decoy structures produced by RosettaAntibody and predict new CDR H3 loop structures de novo. Results When evaluated on the Rosetta antibody benchmark dataset of 49 targets, DeepH3-predicted potentials identified better, same and worse structures [measured by root-mean-squared distance (RMSD) from the experimental CDR H3 loop structure] than the standard Rosetta energy function for 33, 6 and 10 targets, respectively, and improved the average RMSD of predictions by 32.1% (1.4 Å). Analysis of individual geometric potentials revealed that inter-residue orientations were more effective than inter-residue distances for discriminating near-native CDR H3 loops. When applied to de novo prediction of CDR H3 loop structures, DeepH3 achieves an average RMSD of 2.2 ± 1.1 Å on the Rosetta antibody benchmark. Availability and Implementation DeepH3 source code and pre-trained model parameters are freely available at https://github.com/Graylab/deepH3-distances-orientations. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

Geometric Potentials from Deep Learning Improve Prediction of CDR H3 Loop Structures

10.1101/2020.02.09.940254 ◽

2020 ◽

Cited By ~ 3

Author(s):

Jeffrey A. Ruffolo ◽

Carlos Guerra ◽

Sai Pooja Mahajan ◽

Jeremias Sulam ◽

Jeffrey J. Gray

Keyword(s):

Neural Network ◽

Neural Networks ◽

Deep Learning ◽

Deep Neural Networks ◽

Probability Distributions ◽

Loop Structure ◽

Variable Loops ◽

Chain Sequence ◽

Antibody Structure ◽

Complementarity Determining Region

AbstractAntibody structure is largely conserved, except for a complementarity-determining region featuring six variable loops. Five of these loops adopt canonical folds which can typically be predicted with existing methods, while the remaining loop (CDR H3) remains a challenge due to its highly diverse set of observed conformations. In recent years, deep neural networks have proven to be effective at capturing the complex patterns of protein structure. This work proposes DeepH3, a deep residual neural network that learns to predict inter-residue distances and orientations from antibody heavy and light chain sequence. The output of DeepH3 is a set of probability distributions over distances and orientation angles between pairs of residues. These distributions are converted to geometric potentials and used to discriminate between decoy structures produced by RosettaAntibody. When evaluated on the Rosetta Antibody Benchmark dataset of 49 targets, DeepH3-predicted potentials identified better, same, and worse structures (measured by root-mean-squared distance [RMSD] from the experimental CDR H3 loop structure) than the standard Rosetta energy function for 30, 13, and 6 targets, respectively, and improved the average RMSD of predictions by 21.3% (0.48 Å). Analysis of individual geometric potentials revealed that inter-residue orientations were more effective than inter-residue distances for discriminating near-native CDR H3 loop structures.

Download Full-text

DeepSELEX: inferring DNA-binding preferences from HT-SELEX data using multi-class CNNs

Bioinformatics ◽

10.1093/bioinformatics/btaa789 ◽

2020 ◽

Vol 36 (Supplement_2) ◽

pp. i634-i642

Author(s):

Maor Asif ◽

Yaron Orenstein

Keyword(s):

Neural Networks ◽

Dna Binding ◽

High Throughput ◽

Dna Sequences ◽

Deep Neural Networks ◽

High Throughput Sequencing ◽

Supplementary Information ◽

Computational Technique ◽

Model Parameters ◽

Binding Prediction

Abstract Motivation Transcription factor (TF) DNA-binding is a central mechanism in gene regulation. Biologists would like to know where and when these factors bind DNA. Hence, they require accurate DNA-binding models to enable binding prediction to any DNA sequence. Recent technological advancements measure the binding of a single TF to thousands of DNA sequences. One of the prevailing techniques, high-throughput SELEX, measures protein–DNA binding by high-throughput sequencing over several cycles of enrichment. Unfortunately, current computational methods to infer the binding preferences from high-throughput SELEX data do not exploit the richness of these data, and are under-using the most advanced computational technique, deep neural networks. Results To better characterize the binding preferences of TFs from these experimental data, we developed DeepSELEX, a new algorithm to infer intrinsic DNA-binding preferences using deep neural networks. DeepSELEX takes advantage of the richness of high-throughput sequencing data and learns the DNA-binding preferences by observing the changes in DNA sequences through the experimental cycles. DeepSELEX outperforms extant methods for the task of DNA-binding inference from high-throughput SELEX data in binding prediction in vitro and is on par with the state of the art in in vivo binding prediction. Analysis of model parameters reveals it learns biologically relevant features that shed light on TFs’ binding mechanism. Availability and implementation DeepSELEX is available through github.com/OrensteinLab/DeepSELEX/. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

Trigonometric Inference Providing Learning in Deep Neural Networks

Applied Sciences ◽

10.3390/app11156704 ◽

2021 ◽

Vol 11 (15) ◽

pp. 6704

Author(s):

Jingyong Cai ◽

Masashi Takemoto ◽

Yuming Qiu ◽

Hironori Nakajo

Keyword(s):

Machine Learning ◽

Neural Networks ◽

Deep Neural Networks ◽

Activation Function ◽

Trigonometric Approximation ◽

Model Parameters ◽

Training Algorithms ◽

Activation Functions ◽

Classical Training ◽

Sum Formula

Despite being heavily used in the training of deep neural networks (DNNs), multipliers are resource-intensive and insufficient in many different scenarios. Previous discoveries have revealed the superiority when activation functions, such as the sigmoid, are calculated by shift-and-add operations, although they fail to remove multiplications in training altogether. In this paper, we propose an innovative approach that can convert all multiplications in the forward and backward inferences of DNNs into shift-and-add operations. Because the model parameters and backpropagated errors of a large DNN model are typically clustered around zero, these values can be approximated by their sine values. Multiplications between the weights and error signals are transferred to multiplications of their sine values, which are replaceable with simpler operations with the help of the product to sum formula. In addition, a rectified sine activation function is utilized for further converting layer inputs into sine values. In this way, the original multiplication-intensive operations can be computed through simple add-and-shift operations. This trigonometric approximation method provides an efficient training and inference alternative for devices with insufficient hardware multipliers. Experimental results demonstrate that this method is able to obtain a performance close to that of classical training algorithms. The approach we propose sheds new light on future hardware customization research for machine learning.

Download Full-text

MUM&Co: accurate detection of all SV types through whole-genome alignment

Bioinformatics ◽

10.1093/bioinformatics/btaa115 ◽

2020 ◽

Vol 36 (10) ◽

pp. 3242-3243 ◽

Cited By ~ 2

Author(s):

Samuel O’Donnell ◽

Gilles Fischer

Keyword(s):

De Novo ◽

Supplementary Information ◽

Genome Alignment ◽

Whole Genome ◽

Structural Variations ◽

Sequencing Technologies ◽

Third Generation Sequencing ◽

Human Genomes ◽

Whole Genome Alignment ◽

Primary Output

Abstract Summary MUM&Co is a single bash script to detect structural variations (SVs) utilizing whole-genome alignment (WGA). Using MUMmer’s nucmer alignment, MUM&Co can detect insertions, deletions, tandem duplications, inversions and translocations greater than 50 bp. Its versatility depends upon the WGA and therefore benefits from contiguous de-novo assemblies generated by third generation sequencing technologies. Benchmarked against five WGA SV-calling tools, MUM&Co outperforms all tools on simulated SVs in yeast, plant and human genomes and performs similarly in two real human datasets. Additionally, MUM&Co is particularly unique in its ability to find inversions in both simulated and real datasets. Lastly, MUM&Co’s primary output is an intuitive tabulated file containing a list of SVs with only necessary genomic details. Availability and implementation https://github.com/SAMtoBAM/MUMandCo. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

AStrap: identification of alternative splicing from transcript sequences without a reference genome

Bioinformatics ◽

10.1093/bioinformatics/bty1008 ◽

2018 ◽

Vol 35 (15) ◽

pp. 2654-2656 ◽

Cited By ~ 5

Author(s):

Guoli Ji ◽

Wenbin Ye ◽

Yaru Su ◽

Moliang Chen ◽

Guangzao Huang ◽

...

Keyword(s):

Machine Learning ◽

Alternative Splicing ◽

Single Molecule ◽

Reference Genome ◽

De Novo ◽

Supplementary Information ◽

Model Organisms ◽

Sequencing Data ◽

Extensive Evaluation ◽

Reference Genomes

Abstract Summary Alternative splicing (AS) is a well-established mechanism for increasing transcriptome and proteome diversity, however, detecting AS events and distinguishing among AS types in organisms without available reference genomes remains challenging. We developed a de novo approach called AStrap for AS analysis without using a reference genome. AStrap identifies AS events by extensive pair-wise alignments of transcript sequences and predicts AS types by a machine-learning model integrating more than 500 assembled features. We evaluated AStrap using collected AS events from reference genomes of rice and human as well as single-molecule real-time sequencing data from Amborella trichopoda. Results show that AStrap can identify much more AS events with comparable or higher accuracy than the competing method. AStrap also possesses a unique feature of predicting AS types, which achieves an overall accuracy of ∼0.87 for different species. Extensive evaluation of AStrap using different parameters, sample sizes and machine-learning models on different species also demonstrates the robustness and flexibility of AStrap. AStrap could be a valuable addition to the community for the study of AS in non-model organisms with limited genetic resources. Availability and implementation AStrap is available for download at https://github.com/BMILAB/AStrap. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

Motion Capture Sensor-Based Emotion Recognition Using a Bi-Modular Sequential Neural Network

Sensors ◽

10.3390/s22010403 ◽

2022 ◽

Vol 22 (1) ◽

pp. 403

Author(s):

Yajurv Bhatia ◽

ASM Hossain Bari ◽

Gee-Sern Jison Hsu ◽

Marina Gavrilova

Keyword(s):

Deep Learning ◽

Emotion Recognition ◽

Motion Capture ◽

Smart Home ◽

Deep Neural Networks ◽

Research Study ◽

Assistive Robotics ◽

Model Parameters ◽

Modular Architecture ◽

Adaptive Therapy

Motion capture sensor-based gait emotion recognition is an emerging sub-domain of human emotion recognition. Its applications span a variety of fields including smart home design, border security, robotics, virtual reality, and gaming. In recent years, several deep learning-based approaches have been successful in solving the Gait Emotion Recognition (GER) problem. However, a vast majority of such methods rely on Deep Neural Networks (DNNs) with a significant number of model parameters, which lead to model overfitting as well as increased inference time. This paper contributes to the domain of knowledge by proposing a new lightweight bi-modular architecture with handcrafted features that is trained using a RMSprop optimizer and stratified data shuffling. The method is highly effective in correctly inferring human emotions from gait, achieving a micro-mean average precision of 0.97 on the Edinburgh Locomotive Mocap Dataset. It outperforms all recent deep-learning methods, while having the lowest inference time of 16.3 milliseconds per gait sample. This research study is beneficial to applications spanning various fields, such as emotionally aware assistive robotics, adaptive therapy and rehabilitation, and surveillance.

Download Full-text

MPF–BML: a standalone GUI-based package for maximum entropy model inference

Bioinformatics ◽

10.1093/bioinformatics/btz925 ◽

2019 ◽

Vol 36 (7) ◽

pp. 2278-2279

Author(s):

Ahmed A Quadeer ◽

Matthew R McKay ◽

John P Barton ◽

Raymond H Y Louie

Keyword(s):

Maximum Entropy ◽

Fitness Landscape ◽

Sequence Data ◽

Surface Proteins ◽

Supplementary Information ◽

Model Parameters ◽

Maximum Entropy Model ◽

Entropy Model ◽

Widespread Application ◽

Maximum Entropy Models

Abstract Summary Learning underlying correlation patterns in data is a central problem across scientific fields. Maximum entropy models present an important class of statistical approaches for addressing this problem. However, accurately and efficiently inferring model parameters are a major challenge, particularly for modern high-dimensional applications such as in biology, for which the number of parameters is enormous. Previously, we developed a statistical method, minimum probability flow–Boltzmann Machine Learning (MPF–BML), for performing fast and accurate inference of maximum entropy model parameters, which was applied to genetic sequence data to estimate the fitness landscape for the surface proteins of human immunodeficiency virus and hepatitis C virus. To facilitate seamless use of MPF–BML and encourage more widespread application to data in diverse fields, we present a standalone cross-platform package of MPF–BML which features an easy-to-use graphical user interface. The package only requires the input data (protein sequence data or data of multiple configurations of a complex system with large number of variables) and returns the maximum entropy model parameters. Availability and implementation The MPF–BML software is publicly available under the MIT License at https://github.com/ahmedaq/MPF-BML-GUI. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

BICORN: An R package for integrative inference of de novo cis-regulatory modules

Scientific Reports ◽

10.1038/s41598-020-63043-2 ◽

2020 ◽

Vol 10 (1) ◽

Author(s):

Xi Chen ◽

Jinghua Gu ◽

Andrew F. Neuwald ◽

Leena Hilakivi-Clarke ◽

Robert Clarke ◽

...

Keyword(s):

Gene Transcription ◽

Target Genes ◽

De Novo ◽

R Package ◽

Model Parameters ◽

Expression Data ◽

Regulatory Modules ◽

Genome Wide ◽

Context Specific ◽

Tf Gene

Abstract Genome-wide transcription factor (TF) binding signal analyses reveal co-localization of TF binding sites, based on which cis-regulatory modules (CRMs) can be inferred. CRMs play a key role in understanding the cooperation of multiple TFs under specific conditions. However, the functions of CRMs and their effects on nearby gene transcription are highly dynamic and context-specific and therefore are challenging to characterize. BICORN (Bayesian Inference of COoperative Regulatory Network) builds a hierarchical Bayesian model and infers context-specific CRMs based on TF-gene binding events and gene expression data for a particular cell type. BICORN automatically searches for a list of candidate CRMs based on the input TF bindings at regulatory regions associated with genes of interest. Applying Gibbs sampling, BICORN iteratively estimates model parameters of CRMs, TF activities, and corresponding regulation on gene transcription, which it models as a sparse network of functional CRMs regulating target genes. The BICORN package is implemented in R (version 3.4 or later) and is publicly available on the CRAN server at https://cran.r-project.org/web/packages/BICORN/index.html.

Download Full-text

Generalized Born radii computation using linear models and neural networks

Bioinformatics ◽

10.1093/bioinformatics/btz818 ◽

2019 ◽

Vol 36 (6) ◽

pp. 1757-1764

Author(s):

Saida Saad Mohamed Mahmoud ◽

Gennaro Esposito ◽

Giuseppe Serra ◽

Federico Fogolari

Keyword(s):

Neural Network ◽

Neural Networks ◽

Linear Model ◽

Correlation Coefficient ◽

Linear Models ◽

Reference Method ◽

Supplementary Information ◽

Model Parameters ◽

Generalized Born ◽

Supplementary Material

Abstract Motivation Implicit solvent models play an important role in describing the thermodynamics and the dynamics of biomolecular systems. Key to an efficient use of these models is the computation of generalized Born (GB) radii, which is accomplished by algorithms based on the electrostatics of inhomogeneous dielectric media. The speed and accuracy of such computations are still an issue especially for their intensive use in classical molecular dynamics. Here, we propose an alternative approach that encodes the physics of the phenomena and the chemical structure of the molecules in model parameters which are learned from examples. Results GB radii have been computed using (i) a linear model and (ii) a neural network. The input is the element, the histogram of counts of neighbouring atoms, divided by atom element, within 16 Å. Linear models are ca. 8 times faster than the most widely used reference method and the accuracy is higher with correlation coefficient with the inverse of ‘perfect’ GB radii of 0.94 versus 0.80 of the reference method. Neural networks further improve the accuracy of the predictions with correlation coefficient with ‘perfect’ GB radii of 0.97 and ca. 20% smaller root mean square error. Availability and implementation We provide a C program implementing the computation using the linear model, including the coefficients appropriate for the set of Bondi radii, as Supplementary Material. We also provide a Python implementation of the neural network model with parameter and example files in the Supplementary Material as well. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text

Bayesian inference of distributed time delay in transcriptional and translational regulation

Bioinformatics ◽

10.1093/bioinformatics/btz574 ◽

2019 ◽

Vol 36 (2) ◽

pp. 586-593

Author(s):

Boseung Choi ◽

Yu-Yu Cheng ◽

Selahattin Cinar ◽

William Ott ◽

Matthew R Bennett ◽

...

Keyword(s):

Bayesian Inference ◽

Stochastic Systems ◽

Imaging Techniques ◽

Reaction Rates ◽

Biochemical Networks ◽

Supplementary Information ◽

Mcmc Methods ◽

Parameter Estimates ◽

Model Parameters ◽

Cell Functions

Abstract Motivation Advances in experimental and imaging techniques have allowed for unprecedented insights into the dynamical processes within individual cells. However, many facets of intracellular dynamics remain hidden, or can be measured only indirectly. This makes it challenging to reconstruct the regulatory networks that govern the biochemical processes underlying various cell functions. Current estimation techniques for inferring reaction rates frequently rely on marginalization over unobserved processes and states. Even in simple systems this approach can be computationally challenging, and can lead to large uncertainties and lack of robustness in parameter estimates. Therefore we will require alternative approaches to efficiently uncover the interactions in complex biochemical networks. Results We propose a Bayesian inference framework based on replacing uninteresting or unobserved reactions with time delays. Although the resulting models are non-Markovian, recent results on stochastic systems with random delays allow us to rigorously obtain expressions for the likelihoods of model parameters. In turn, this allows us to extend MCMC methods to efficiently estimate reaction rates, and delay distribution parameters, from single-cell assays. We illustrate the advantages, and potential pitfalls, of the approach using a birth–death model with both synthetic and experimental data, and show that we can robustly infer model parameters using a relatively small number of measurements. We demonstrate how to do so even when only the relative molecule count within the cell is measured, as in the case of fluorescence microscopy. Availability and implementation Accompanying code in R is available at https://github.com/cbskust/DDE_BD. Supplementary information Supplementary data are available at Bioinformatics online.

Download Full-text