Rohit Singh

COMPASS: Component-Wise Inference of Shared and Gene-Specific Perturbation Response

The paper introduces COMPASS, a model that decomposes genetic perturbation responses into shared and gene-specific components, showing improved accuracy and discrimination compared to existing methods.

Foundation model reveals the shared organization of transcription and topologically associating domains

The paper develops a foundation model to analyze TADs and their impact on transcription, finding that TADs enhance co-regulation and this effect declines with aging and changes in cancer cells.

TP53-mediated bidirectional lineage plasticity drives alveolar epithelial cell extrusion and tissue remodeling

The paper identifies TP53 activation in alveolar epithelial cells as a driver of cell extrusion and tissue remodeling in IPF through bidirectional lineage plasticity.

Decoding the causal drivers of spatial cellular topology

The paper introduces GLACIER, a method that infers spatially encoded transcriptional and signaling relationships using spatial Granger causality, applied to mouse cerebellum single-cell spatial transcriptomics data.

Local transcriptional covariation produces accurate estimates of cell phenotype

This paper introduces Sceodesic, a method that improves cell phenotype estimation from scRNA-seq by analyzing local gene coexpression patterns using Riemannian geometry, outperforming existing approaches in prediction and biological interpretation.

The Immunoregulatory Architecture of the Adult Oral Cavity

This paper presents a single-cell and spatial proteotranscriptomic atlas of human oral tissues, identifying fibroblasts as key immunoregulatory hubs and revealing changes in disease states.

EvoPool: Evolution-Guided Pooling of Protein Language Model Embeddings

EvoPool improves variant effect prediction by integrating evolutionary information into protein language model embeddings using optimal transport.

DeltaNMF: A Two-Stage Neural NMF for Differential Gene Program Discovery

DeltaNMF improves differential gene program discovery in case-control studies by separating baseline and case-specific programs using a two-stage neural NMF framework.

Learning a PRECISE language for small-molecule binding

PRECISE improves virtual screening by combining quantized drug representations with 3D target structures to predict binding sites and enable billion-scale screening.

USHER: Guiding Foundation Model Representations through Distribution Shifts

USHER corrects out-of-distribution biases in foundation models by transforming embeddings to a reference space, improving cross-platform integration and downstream tasks.

Iterative immunogen optimization to focus elicited responses on a conserved, subdominant influenza hemagglutinin epitope

The paper designs immunogens that redirect immune responses to a conserved influenza epitope using iterative machine learning and in vivo testing.

Democratizing Protein Language Model Training, Sharing and Collaboration

The paper introduces SaprotHub, a platform that simplifies training and sharing of protein language models, enabling non-experts to build and collaborate on customized models.

Unveiling causal regulatory mechanisms through cell-state parallax

The paper introduces GrID-Net, a graph neural network that detects causal locus-gene associations using cell-state parallax, uncovering noncoding mechanisms affecting 132 genes in schizophrenia.

SAME: Topology-flexible transforms enable robust integration of multimodal spatial omics

SAME improves multimodal spatial omics integration by allowing topological disruptions during alignment, enhancing cell-type accuracy and revealing hidden biological insights.

Deep Learning-coupled Proximity Proteomics to Deconvolve Kinase Signaling In Vivo

The paper introduces KolossuS, a deep learning framework for kinase-substrate prediction, integrated with CRISPR-based proximity proteomics to identify kinase substrates in vivo, including novel substrates associated with sleep deprivation.

The basic helix-loop-helix transcription factor TCF4 recruits the Mediator Complex to activate gonadal genes and drive ovarian development

This study identifies TCF4 as a key regulator of ovarian development by recruiting the Mediator Complex to activate gonadal genes and suppress male-specific pathways.

Topology-Driven Discovery of Transmembrane Protein S-Palmitoylation

The paper develops a machine learning tool to predict S-palmitoylation sites on transmembrane proteins, identifying new candidates and demonstrating its utility in design.

Learning a CoNCISE language for small-molecule binding

This paper introduces CoNCISE, a method that accelerates drug-target interaction prediction by compressing small molecule data into 15 bits per compound, achieving high accuracy and enabling rapid screening of large chemical libraries.

Decoding the Functional Interactome of Non-Model Organisms with PHILHARMONIC

PHILHARMONIC infers protein interaction networks and functional organization in non-model organisms using deep learning and spectral clustering, validated in Drosophila and coral symbionts.

Learning the Language of Antibody Hypervariability

This paper introduces AbMAP, a transfer learning framework for antibody sequence analysis, showing it can predict structure, binding, and paratope with high accuracy and enable antibody optimization.

Aggregating Residue-Level Protein Language Model Embeddings with Optimal Transport

This paper introduces an optimal transport-based method to aggregate residue-level protein language model embeddings into fixed-length representations, showing improved performance over average pooling for downstream tasks.

Allo-Allo: Data-efficient prediction of allosteric sites

Allo-Allo predicts allosteric sites using a data-efficient sequence-based method, achieving higher performance than existing approaches and showing translational potential in disease risk association.

Miniaturizing, Modifying, and Magnifying Nature's Proteins with Raygun

Raygun is a generative AI framework that miniaturizes, modifies, and expands proteins while preserving function and structural integrity, validated through cell-based experiments.

Comparative analysis of the syncytiotrophoblast in placenta tissue and trophoblast organoids using snRNA sequencing

This study compares syncytiotrophoblast gene expression in placental tissue and trophoblast organoids using snRNA-seq, identifying distinct nuclear subtypes and environmental influences on STB gene expression.

Causal gene regulatory analysis with RNA velocity reveals an interplay between slow and fast transcription factors

Velorama infers causal gene regulatory networks using RNA velocity data, revealing TF interaction speeds and their links to disease.

Split-intein Gal4 provides intersectional genetic labeling that is fully repressible by Gal80

This paper introduces a new split-Gal4 system based on a self-excising split-intein that is fully repressible by Gal80, enabling temporal control of genetic labeling in Drosophila.

Transfer of Knowledge from Model Organisms to Evolutionarily Distant Non-Model Organisms: The Coral Pocillopora damicornis Membrane Signaling Receptome

The paper develops a method to identify homologous membrane receptors in non-model corals by combining remote homology detection and functional analysis, mapping over 1000 human receptors to coral receptors.

Learning the Drug-Target Interaction Lexicon

ConPLex predicts drug-target interactions using a deep learning model that combines pre-trained protein language models with contrastive co-embedding, achieving high accuracy and scalability.

Adapting protein language models for rapid DTI prediction

This paper presents a deep learning approach using pre-trained protein language models for drug-target interaction prediction, achieving higher accuracy and faster training than existing methods.

Contrasting drugs from decoys

This paper introduces CON-Plex, a drug-target interaction model that improves specificity by distinguishing drugs from decoys using contrastive learning.