Biological representation disentanglement of single-cell data

Abstract

Due to its internal state or external environment, a cell’s gene expression profile contains multiple signatures, simultaneously encoding information about its characteristics. Disentangling these factors of variations from single-cell data is needed to recover multiple layers of biological information and extract insight into the individual and collective behavior of cellular populations. While several recent methods were suggested for biological disentanglement, each has its limitations; they are either task-specific, cannot capture inherent nonlinear or interaction effects, cannot integrate layers of experimental data, or do not provide a general reconstruction procedure. We presentbiolord, a deep generative framework for disentangling known and unknown attributes in single-cell data. Biolord exposes the distinct effects of different biological processes or tissue structure on cellular gene expression. Based on that, biolord allows generating experimentally-inaccessible cell states by virtually shifting cells across time, space, and biological states. Specifically, we showcase accurate predictions of cellular responses to drug perturbations and generalization to predict responses to unseen drugs. Further, biolord disentangles spatial, temporal, and infection-related attributes and their associated gene expression signatures in a single-cell atlas ofPlasmodiuminfection progression in the mouse liver. Biolord can handle partially labeled attributes by predicting a classification for missing labels, and hence can be used to computationally extend an infected hepatocyte population identified at a late stage of the infection to earlier stages. Biolord applies to diverse biological settings, is implemented using the scvi-tools library, and is released as open-source software athttps://github.com/nitzanlab/biolord.

Signals in the Cells: Multimodal and Contextualized Machine Learning Foundations for Therapeutics

“…Code for intra/inter cell-line benchmarks for chemical (drug) and genetic (CRISPR) perturbations is Table 2: We've evaluated chemCPA utilizing cold splits on perturbation type and show a significant decrease in performance for 3 of 4 perturbations evaluated. We've also included Biolord [42] and scGen [91] for comparison. The dataset used was 4 chemical (drug) perturbations from sciPlex2 [2].…”

“…TDC-2 introduces three new learning tasks focusing on cell-type-specific biological contexts, drug-target identification [3], and prediction of responses to chemical and genetic perturbations [20, 19, 42]. TDC-2 is the first renowned multimodal open-source dataset and benchmark provider to introduce a protein-peptide binding affinity prediction task [9] and a precision-medicine-oriented clinical trial outcome prediction task [8].…”

“…TDC-2 provides benchmarks for over 15 state-of-the-art models across more than five new learning tasks. These are tailored to take on some of the most pressing machine learning challenges in biomedicine, including but not limited to cell-type-specific machine learning modeling and evaluation [3], the inferential gap in precision medicine [41], negative-sampling challenges in peptidomimetics [22], and out-of-distribution model generalizability across unseen cell lines and perturbations [42, 43].…”

“…TDC-2 introduces a protein-peptide binding affinity prediction task [9] and clinical trial outcome prediction task [8], altogether providing tasks across antigen-processingpathway contexts, cell type contexts, and patient contexts. TDC-2 is designed to support ML research in some of the most pressing challenges, including but not limited to cell-type-specific ML modeling [3], inferential gap in precision medicine [41], negativesampling challenges in peptidomimetics [22], and OOD generalization in perturbations [42,43]. TDC-2 is focused on providing functionality to support therapeutic foundation model research.…”

“…TDC-2 is designed to support ML research in some of the most pressing challenges, including but not limited to cell-type-specific ML modeling [3], inferential gap in precision medicine [41], negative-sampling challenges in peptidomimetics [22], and OOD generalization in perturbations [42, 43]. TDC-2 is focused on providing functionality to support therapeutic foundation model research.…”