Biocomputing 2022 - Proceedings Of The Pacific Symposium.

• Author: Altman, Russ B.
• TeilnehmendeR: Dunker, A Keith.; Hunter, Lawrence.; Ritchie, Marylyn D.; Murray, Tiffany A.; Klein, Teri E.
• Place / Publishing House: Singapore : : World Scientific Publishing Company,, 2021.; Ã2022.
• Year of Publication: 2021
• Edition: 1st ed.
• Language: English
• Physical Description: 1 online resource (431 pages)

Table of Contents

• Intro
• Content
• Preface
• AI-DRIVEN ADVANCES IN MODELING OF PROTEIN STRUCTURE
• Session Introduction: AI-Driven Advances in Modeling of Protein Structure
• 1. A short retrospect
• 2. A brief outline of current research
• 3. Future developments (complexes, ligand interactions, other molecules, dynamics, language models, geometry models, sequence design)
• 4. What is needed for further progress?
• 5. Overview of papers in this session
• 5.1. Evaluating significance of training data selection in machine learning
• 5.2. Geometric pattern transferability
• 5.3. Supervised versus unsupervised sequence to contact learning
• 5.4. Side chain packing using SE(3) transformers
• 5.5. Feature selection in electrostatic representations of ligand binding sites
• References
• Training Data Composition Affects Performance of Protein Structure Analysis Algorithms
• 1. Introduction
• 2. Methods
• 2.1. Experimental Design
• 2.2. Task-specific Methods
• 3. Results
• 3.1. Performance on NMR and cryo-EM structures is consistently lower than performance on X-ray structures, independent of training set
• 3.2. Inclusion of NMR data in the training set improves performance on held-out NMR data and does not degrade performance on X-ray data
• 3.3. Known biochemical and biophysical effects are replicated in trained models
• 3.4. Downsampling X-ray structures during training negatively affects performance on all types of data
• 4. Conclusion
• 5. Acknowledgments
• References
• Transferability of Geometric Patterns from Protein Self-Interactions to Protein-Ligand Interactions
• 1. Introduction
• 2. Related Work
• 3. Methods
• 3.1. Datasets
• 3.2. Contact extraction
• 3.3. Representing contact geometry
• 4. Results
• 4.1. Protein self-contacts exhibit clear geometric clustering.
• 4.2. Many geometric patterns transfer to protein-ligand contacts
• 4.3. Application to protein-ligand docking
• 5. Conclusion and Future Work
• Supplemental Material, Code, and Data Availability
• Acknowledgments
• References
• Interpreting Potts and Transformer Protein Models Through the Lens of Simplified Attention
• 1. Introduction
• 2. Background
• 3. Methods
• 3.1. Potts Models
• 3.2. Factored Attention
• 3.3. Single-layer attention
• 3.4. Pretraining on Sequence Databases
• 3.5. Extracting Contacts
• 4. Results
• 5. Discussion
• Acknowledgements
• References
• Side-Chain Packing Using SE(3)-Transformer
• 1. Introduction
• 2. Methods
• 2.1. Neighborhood Graph Representation
• 2.2. The SE(3)-Transformer Architecture
• 2.3. Node Features
• 2.4. Final Layer
• 2.5. Rotamer Selection
• 2.6. Experiments
• 3. Results
• 4. Conclusion
• 5. Acknowledgements
• 6. References
• DeepVASP-E: A Flexible Analysis of Electrostatic Isopotentials for Finding and Explaining Mechanisms that Control Binding Specificity
• 1. Introduction
• 2. Methods
• 2.1. Convolutional Neural Network
• 2.2. Experimental Design
• 2.3. Comparison with Existing Methods
• 3. Results
• 4. Conclusions
• Acknowledgements
• References
• BIG DATA IMAGING GENOMICS
• Session Introduction: Big Data Imaging Genomics
• 1. Introduction
• 2. Overview of Contributions
• References
• A New Mendelian Randomization Method to Estimate Causal Effects of Multivariable Brain Imaging Exposures
• 1. Introduction
• 2. Methods
• 2.1. Step 1 : Mendelian randomization analysis on a single imaging exposure
• 2.2. Step 2: Joint instrumental variables and imaging exposures selection
• 2.3. Step 3: Causal effect identification for multiple imaging exposures
• 3. Application to evaluate the causal effect of white matter microstructure integrity on cognitive function.
• 3.1. Data and study cohort
• 3.2. Results
• 4. Simulation
• 5. Discussion
• Funding
• Availability of data and materials
• Authors' contributions
• References
• Efficient Differentially Private Methods for a Transmission Disequilibrium Test in Genome Wide Association Studies
• 1. Introduction
• 2. Preliminaries
• 2.1. TDT
• 2.2. Differential Privacy
• 3. Methods
• 3.1. Exact Algorithm
• 3.2. Approximation Algorithm
• 4. Experiments
• 4.1. Simulation Data
• 4.2. Results
• 4.2.1. Run Time
• 4.2.2. Accuracy
• 4.3. Real Data
• 5. Conclusion
• Acknowledgement
• References
• Identifying Imaging Genetic Associations via Regional Morphometricity Estimation
• 1. Introduction
• 2. Methods
• 3. Materials
• 4. Experimental Design
• 5. Results and Discussion
• 6. Conclusion
• Acknowledgements
• References
• Identifying Highly Heritable Brain Amyloid Phenotypes Through Mining Alzheimer's Imaging and Sequencing Biobank Data
• 1. Introduction
• 2. Method
• 3. Materials
• 4. Experimental Workow
• 5. Results and Discussion
• 6. Conclusion
• Acknowledgements
• References
• Effects of ApoE4 and ApoE2 Genotypes on Subcortical Magnetic Susceptibility and Microstructure in 27,535 Participants from the UK Biobank
• 1. Introduction
• 2. Methods
• 2.1. UK Biobank Participants
• 2.2. T1-Weighted MRI
• 2.3. Quantitative Magnetic Susceptibility
• 2.4. Diffusion-Weighted MRI
• 2.5. Statistical Analyses
• 3. Results
• 3.1. ApoE4 Microstructural Associations
• 3.2. ApoE2 Microstructural Associations
• 3.3. ApoE-by-Age Interactions
• 3.3.1. ApoE Associations Stratified by Age
• 4. Discussion
• References
• Separating Clinical and Subclinical Depression by Big Data Informed Structural Vulnerability Index and Its impact on Cognition: ENIGMA Dot Product
• 1. Introduction
• 2. Methods.
• 2.1 Participants.
• 2.2 Major Depressive Disorder Classification
• 2.3 Imaging Protocol and Processing
• 2.4 Calculation of linear indices of similarity
• 2.5 Calculation of QRI
• 2.7 Cognitive assessment
• 2.8 Statistics
• 3. Results
• 3.1 Group differences in symptoms and biomarkers
• 3.2 Effects of MDD on cognition.
• 3.3. Cognitive association
• 4. Discussion.
• 5. Conclusion
• 6. Acknowledgement
• References
• Generalizing Few-Shot Classification of Whole-Genome Doubling Across Cancer Types
• 1. Introduction
• 2. Related Work
• 3. Cohort
• 3.1. Cohort Selection
• 3.2. Feature Extraction
• 4. Methods
• 4.1. Model
• 4.2. Training
• 4.2.1. Pre-Training
• 4.2.2. Meta-Training
• 4.3. Meta-Validation and Meta-Test
• 4.4. Experiments
• 4.4.1. Cancer Types
• 4.4.2. Batch Effects
• 5. Results
• 5.1. Cancer Types
• 5.2. Batch Effects
• 5.2.1. Image Resolution
• 5.2.2. Image Brightness
• 6. Discussion
• Software and Data
• References
• HUMAN INTRIGUE: META-ANALYSIS APPROACHES FOR BIG QUESTIONS WITH BIG DATA WHILE SHAKING UP THE PEER REVIEW PROCESS
• Session Introduction: Human Intrigue: Meta-Analysis Approaches for Big Questions with Big Data While Shaking Up the Peer Review Process
• 1. Introduction
• 2. The Crowd Peer Review Process
• 2.1 Reviewer's Feedback
• 2.2 Conclusions
• 3. Meta-Analysis in Biocomputing
• 3.1 Novel Methods for Meta-Analysis of 'Omics Data
• 3.2 Using Publicly Available Data in Methods Development
• 3.3 Studying the Structure of Publicly Available Data
• 3.4 Conclusions
• Acknowledgements
• References
• Multitask Group Lasso for Genome Wide Association Studies in Diverse Populations
• 1. Introduction
• 2. Methods
• 2.1. Population stratification
• 2.2.1. Adjacency-constrained hierarchical clustering
• 2.2.2. LD-groups across populations
• 2.3. Multitask group Lasso.
• 2.3.1. General framework and problem formulation
• 2.3.2. Related work
• 2.3.3. Gap safe screening rules
• 2.4. Stability selection
• 3. Experiments
• 3.1. Data
• 3.2. Preprocessing
• 3.3. Comparison partners
• 4. Results
• 4.1. MuGLasso draws on both LD-groups and the multitask approach to recover disease SNPs
• 4.2. MuGLasso provides the most stable selection
• 4.3. MuGLasso selects both task-speci c and global LD-groups
• 5. Discussion and Conclusions
• Acknowledgments
• Supplementary Materials and code
• References
• Mixed Effects Machine Learning Models for Colon Cancer Metastasis Prediction Using Spatially Localized Immuno-Oncology Markers
• 1. Introduction
• 2. Motivation for Comparison Study
• 2.1. Review of Prior Spatial Omics Analysis Methods
• 2.2. Motivation for Mixed Effects Machine Learning Approaches
• 3. Materials and Methods
• 3.1. Data Acquisition and Preprocessing
• 3.2. Experimental Design: Prediction Tasks and Modeling Approaches
• 4. Results
• 4.1. Macro: Inter-Tumoral Prediction
• 4.2. METS: Nodal and Distant Metastasis Prediction
• 5. Discussion
• 6. Conclusion
• 7. Acknowledgements
• 8. References
• Improving QSAR Modeling for Predictive Toxicology Using Publicly Aggregated Semantic Graph Data and Graph Neural Networks
• 1. Introduction
• 2. Methods
• 2.1. Obtaining toxicology assay data
• 2.2. Aggregating publicly available multimodal graph data
• 2.3. Heterogeneous graph neural network
• 2.3.1. Node classification
• 2.4. Baseline QSAR classifiers
• 3. Results
• 3.1. GNN node classification performance vs. baseline QSAR models
• 3.2. Ablation analysis of graph components' inuence on the trained model
• 4. Discussion
• 4.1. GNNs versus traditional ML for QSAR modeling
• 4.2. Interpretability of GNNs in QSAR
• 4.3. Sources of bias and their effects on QSAR for toxicity prediction.
• 5. Conclusions.