scgpt

A framework for state-of-the-art pre-trained bio foundation models on genomics and transcriptomics modalities.

Recovering gene regulatory networks from Perturb-seq by fitting steady-state ODEs through frozen single-cell foundation models (GenBio @ ICML 2026)

PhenoSeq: Conditional Diffusion Model; Generating Transcriptomic Representations from Cell Painting Images

Code, results and paper for "The geometry of single-cell foundation models: what they inherit, what they add, and what shapes it"

Causal token interventions for validating gene regulatory network inference from single-cell foundation models

A drop-in modernization of bowang-lab/scGPT for Python 3.12 + torch 2.6 + flash-attn 3 (H100 sm_90a native). Original pretrained weights load unmodified — compatible, more modern, faster.

Patient-aware benchmark of frozen scGPT versus Scanpy/PCA on 29,614 NSCLC single cells with held-out evaluation and sample-level analysis

Companion code for the PLOS ONE paper 'What topological and geometric structure do biological foundation models learn? Evidence from 141 hypotheses'.

Benchmarks four single-cell foundation models: scGPT, UCE, SCimilarity, and Transcriptformer in a zero-shot setting across four downstream tasks:

Автосервис в Минске. Качественный ремонт авто.

Predict transcriptional responses to genetic perturbations, evaluated against baselines that are hard to beat.

scGPT fine-tuning on PBMC3k · 7-method benchmark · Nature Methods 2024 framework

Standardized benchmark for bulk RNA-seq deconvolution: 73 methods x 42 datasets (30 pseudo-bulk + 12 real-bulk). Classical regression, container methods, single-cell foundation-model embeddings, scGPT-LoRA, and ML baselines.

Do cell states from single-cell foundation models (scGPT, Geneformer) predict breast-cancer survival better than an HVG-PCA baseline? Tested in TCGA-BRCA against permutation and random-gene controls, with a frozen-signature replication in METABRIC.

Reproducible pipelines comparing bulk RNA-seq, single-cell (pseudobulk), and scFM embeddings (scGPT, scFoundation) for per-drug response prediction in cancer cell lines.

BG Internship 2026 Group 7: scGPT embeddings, cAE batch correction, strict 35-month OS prediction (v5 pipeline).

CytoBridge: single-cell perturbation prediction with pathway-gated cross-attention; frozen benchmark for 'Beyond per-pair correlation: a drug-discrimination control exposes prediction collapse in single-cell multi-drug perturbation models'