#1 · Systems biology
Deep learning extends variant, regulatory and chromatin interaction prediction
The dominant thread is the augmentation of sequence-based prediction with orthogonal molecular evidence. MPRabc combines MPRA-predicted reporter activity, sequence features, epigenomic signals and 3D contacts to improve enhancer-gene prediction, while SCEG-HiC uses bulk Hi-C as a prior for single-cell enhancer-gene links and GraphChIAr augments Hi-C maps with ChIP-seq profiles. MOSAIC fuses a DNA language model, multi-scale CNN and annotation features for noncanonical splice-altering variants, and Porcine MutBERT introduces SNP-targeted masking in lightweight genomic foundation models. A multi-species benchmark of ten splicing predictors reports AUROC losses of roughly 21 to 33 percent for task-specific and foundation models. Other items extend this pattern to causal network inference grounded in molecular evidence, base-resolution INDEL mutation rate maps, and long-read structural variant calling. What is new is the explicit integration of heterogeneous experimental layers rather than sequence alone.
Published papers
SiSaNA: A Python-based command line interface for Single-Sample Network Analysis.
SiSaNA is a Python command-line workflow for single-sample regulatory network reconstruction and analysis, producing summary HTML reports without programming…
dictyBase 2026: a reimplementation of the Dictyostelium model organism database on minimal infrastructure.
Reimplementation of the dictyBase Dictyostelium model organism database as a static-file service with gene records, sequence search, enrichment tools, stock…
PoolParty: streamlined design of DNA sequence libraries in Python.
PoolParty, a Python package that builds oligo pool DNA libraries as computational graphs, with 50+ mutagenesis, motif, and barcode operations plus design cards…
Preprints
Asking the Data: Conversational Interrogation of LC-MS/MS Proteomics Data Sets with an LLM Agent
Introduces a conversational LLM-agent framework for interrogating LC-MS/MS proteomics datasets, with executable checks, controls, and traceable decisions;…
Benchmarking Large Language Models for Transcriptomic Data Analysis and Visualization: Multi-Model Workflow Evaluation
Benchmarks LLMs and agentic coding systems on automated transcriptomic analysis and visualization of COVID-19 host responses, finding prompt specificity aids…
BaitBench: a comprehensive tool for building, as-sessing, and predicting outcomes for hybrid capture enrichment probes
BaitBench is a Rust tool unifying hybrid capture probe design, assessment, and thermodynamic simulation, with CLI and GUI access; correlation rho=0.916 against…
GLESKit: an MCP-based system for gene-level biomedical investigation and evidence synthesis
GLESKit is an MCP-based system linking gene-level databases, tools and literature into retained workspaces for reviewable reports, evaluated on GeneReportBench…
ProtoPure: Literature Mining and Analysis of Protein Purification Protocols at Proteome Scale
ProtoPure: an LLM-based workflow and database extracting 12 standardized purification parameters from literature across UniProt proteomes, revealing…
tWhEEL: a transcript-aware workflow for quantifying CRISPR knockout feasibility using transcriptome assembly
tWhEEL is a transcript-aware pipeline that infers gene redundancy from transcripts, pairs it with CRISPR guide designs, and scores knockout feasibility across…
Predicting Differential Protein-Protein Interactions
Random forest model using MINT embeddings and condition-specific mass-spec abundance to predict differential protein-protein interactions across eight cell…
Topic history
| Date | Rank |
|---|---|
| #1 | |
| #5 | |
| #4 | |
| #4 | |
| #3 | |
| #4 |