Modeling gene expression cascades during cell state transitions

During cellular processes such as differentiation or response to external stimuli, cells exhibit dynamic changes in their gene expression profiles. Single-cell RNA sequencing (scRNA-seq) can be used to investigate these dynamic changes. To this end, cells are typically ordered along a pseudotemporal trajectory which recapitulates the progression of cells as they transition from one cell state to another. We infer transcriptional dynamics by modeling the gene expression profiles in pseudotemporally ordered cells using a Bayesian inference approach. This enables ordering genes along transcriptional cascades, estimating differences in the timing of gene expression dynamics, and deducing regulatory gene interactions. Here, we apply this approach to scRNA-seq datasets derived from mouse embryonic forebrain and pancreas samples. This analysis demonstrates the utility of the method to derive the ordering of gene dynamics and regulatory relationships critical for proper cellular differentiation and maturation across a variety of developmental contexts.

Distance-weighted Sinkhorn loss for Alzheimer's disease classification

Traditional loss functions such as cross-entropy loss often quantify the penalty for each mis-classified training sample without adequately considering its distance from the ground truth class distribution in the feature space. Intuitively, the larger this distance is, the higher the penalty should be. With this observation, we propose a penalty called distance-weighted Sinkhorn (DWS) loss. For each mis-classified training sample (with predicted label A and true label B), its contribution to the DWS loss positively correlates to the distance the training sample needs to travel to reach the ground truth distribution of all the A samples. We apply the DWS framework with a neural network to classify different stages of Alzheimer's disease. Our empirical results demonstrate that the DWS framework outperforms the traditional neural network loss functions and is comparable or better to traditional machine learning methods, highlighting its potential in biomedical informatics and data science.

Classification of Neisseria meningitidis genomes with a bag-of-words approach and machine learning

Whole genome sequencing of bacteria is important to enable strain classification. Using entire genomes as an input to machine learning (ML) models would allow rapid classification of strains while using information from multiple genetic elements. We developed a "bag-of-words" approach to encode, using SentencePiece or k-mer tokenization, entire bacterial genomes and analyze these with ML. Initial model selection identified SentencePiece with 8,000 and 32,000 words as the best approach for genome tokenization. We then classified in Neisseria meningitidis genomes the capsule B group genotype with 99.6% accuracy and the multifactor invasive phenotype with 90.2% accuracy, in an independent test set. Subsequently, in silico knockouts of 2,808 genes confirmed that the ML model predictions aligned with our current understanding of the underlying biology. To our knowledge, this is the first ML method using entire bacterial genomes to classify strains and identify genes considered relevant by the classifier.

SuSiE PCA: A scalable Bayesian variable selection technique for principal component analysis

Latent factor models, like principal component analysis (PCA), provide a statistical framework to infer low-rank representation in various biological contexts. However, feature selection is challenging when this low-rank structure manifests from a sparse subspace. We introduce SuSiE PCA, a scalable sparse latent factor approach that evaluates uncertainty in contributing variables through posterior inclusion probabilities. We validate our model in extensive simulations and demonstrate that SuSiE PCA outperforms other approaches in signal detection and model robustness. We apply SuSiE PCA to multi-tissue expression quantitative trait loci (eQTLs) data from GTEx v8 and identify tissue-specific factors and their contributing eGenes. We further investigate its performance on the large-scale perturbation data and find that SuSiE PCA identifies modules with a higher enrichment of ribosome-related genes than sparse PCA (false discovery rate [FDR] = 9.2 × 10 - 82 vs. 1.4 × 10 - 33 ), while being ∼ 18x faster. Overall, SuSiE PCA provides an efficient tool to identify relevant features in high-dimensional biological data.