perturbation screen

Designed protein binders to conserved MYC regions reach low-nanomolar affinity, reduce MYC-MAX proximity in cells, and are depleted from MYC-dependent cancer cells in culture and mouse pancreatic tumors.

Predicting differentially expressed genes and expression shifts per condition from database embeddings plus per-gene evidence, PIE reaches 1.2–3.2 times the best baseline's AUPRC on unseen contexts, perturbations and datasets.

A set-aware latent transport model predicts treated cell populations from unpaired controls, keeps perturbation-specific signal in CRISPR, drug and cytokine data, and guided a cytokine choice tested in three donors.

Across 13 single-cell foundation models and 94 datasets, PCA and established methods usually match or beat them; pretrained embeddings help mainly with scarce labels, rare cells and open-set detection.

State, a set-based transformer trained on over 100 million perturbed cells, predicts perturbation responses in new cellular contexts better than mean, linear and deep learning baselines.

Genome-scale CRISPRi perturb-seq in 22 million primary human CD4+ T cells maps stimulation-specific regulators and links perturbation signatures to T cell states and autoimmune genetics.

STITCH pairs size-independent tangent-PCA contour shape modes with expression factors by canonical correlation, recovering keratinocyte, fibroblast and melanoma shape-expression links in Xenium and CosMx data.

A >50,000-element viral smORF library screened by growth, MHC-I, ER stress, Perturb-seq and mass spectrometry yields >2,000 active microproteins and shows EBV BNLF2b inhibits PKA by mimicry.

Energy-model maps of five human kinases show conserved distance-dependent decay of inhibitory allostery, but distal allosteric pockets and activating sites diverge between kinases.

DRiF-Seq, a sequencing-optimized RNAi platform with a posteriori noise estimation, quantifies clone-resolved fitness genome-wide in pleomorphic Trypanosoma brucei; over half of core genes carry fitness costs.

Proteogenomic analysis of up to 11 CPTAC cohorts shows ubiquitin-proteasome proteins often diverge from transcripts; TP53-mutant tumors elevate UBR5 protein, while TRIM28 prognosis is lineage-dependent.

A CRISPRi migration screen in neutrophil-like HL-60 cells identified Galvanin (TMEM154), a single-pass membrane protein whose negatively charged ectodomain electrophoreses anodally to steer cathodal electrotaxis.

Cell Painting in primary human hepatocytes exposed to 1,085 compounds detects bioactivity at lower concentrations than cytotoxicity assays and predicts cell-based, but not cell-free, ToxCast endpoints across feature extractors.

A HaloTagged MFN2 reporter enabled CRISPRi screens showing diverse mitochondrial damage activates PINK1-Parkin via membrane potential loss, with TOM retaining PINK1 when TIM23 is blocked.

A 100-million-cell atlas: 50 cancer lines pooled as spheroid villages, treated with 379 drugs at several doses, profiled by combinatorial-barcoding scRNA-seq and embedded with a 10-dimensional scVI model.

Consensus likelihoods from general, unsupervised protein language models evolve seven human antibodies using only a wildtype sequence and <=20 tested variants, improving binding affinity up to 160-fold.

Cell Painting images and profiles for 115,796 compounds in U2OS cells from 12 laboratories; 15.2% of analyzed compounds are phenotypically active, and cross-laboratory replicates cut reference-compound activity by 34%.

MOFA, a sparse Bayesian factor model with mixed likelihoods, separates shared from modality-specific multi-omic variation, recovering IGHV, trisomy 12 and oxidative-stress axes in 200 leukaemia samples.

In suspended human keratinocytes, five induced phosphatases dampen ERK to drive differentiation commitment, opposed by DUSP10, in a time-rewired network that renders commitment an unstable switch.