> 매일 아침 research-watcher가 27개 query family × PubMed/GEO를 스캔해서, Brown Biotech 파이프라인 관점에서 decision-ready 인사이트로 정리합니다. 매일 06:00 KST 자동 생성. **Category focus:** AI Drug Discovery **Published:** 2026-08-14 06:09 KST **Entry ID:** 68 **Tags:** #gse343370 #gse343371 #mskcc #dual-barcoding #pooled-crispr-ko #30ko #gem-x #perturbation-atlas #scrna-seq #single-cell #h5ad #ai-drug-discovery --- ## 🔬 Today's Top Findings ### 1. MSKCC dual-barcoding pooled KO atlas (GSE343370+GSE343371, score 9, mixed + single-cell, n=3+n=24, pdat 2026/08/12 = 1 day ago, the FRESHEST score-9 dual-barcoding pooled CRISPR-KO atlas in the entire 2026-08-14 scan and a direct AI-DD + bioinformatics + perturbation-omics input for the BB AI-DD lane) ### 2. Integrated single-cell + spatial ovarian immune landscape (GSE308275, score 9, n=6, pdat 2026/08/12 = 1 day ago, the FRESHEST score-9 ovarian-cancer + single-cell + spatial immune-microenvironment atlas in the 2026-08-14 scan and a direct refractory-ovarian-cancer + AI-DD input for the BB bioinformatics + AI-DD lane) ### 3. Longitudinal single-cell + spatial intratumor heterogeneity atlas (GSE343503, score 9, n=7, pdat 2026/08/12 = 1 day ago, the FRESHEST score-9 longitudinal + single-cell + spatial intratumor-heterogeneity atlas in the 2026-08-14 scan and a direct refractory-cancer + AI-DD input for the BB bioinformatics + AI-DD lane) ## 📋 Synthesis The 2026-08-14 06:00 KST research-watcher scan completed with 105 hits across 27 query families and yielded 4 FRESH score-9 NOVEL accessions (all pdat 2026/08/12 = 1 day ago, all absent from the prior id:62-id:67 window at both accession and study-family level), recovering cleanly after yesterday's id:67 mixed-recovery day. The four fresh score-9 NOVEL entries: GSE343370 + GSE343371 (MSKCC dual-barcoding pooled 30KO + 30KO_GEM-X CRISPR-perturbation + scRNA-seq companion set, n=3 + n=24, H5AD raw files) — a perturbation-omics + AI-DD training-data goldmine for pooled-screen target nomination; GSE308275 (integrated single-cell + spatial ovarian-cancer immune landscape, n=6, 1 day fresh) — a refractory-ovarian-cancer immune-microenvironment atlas; GSE343503 (longitudinal single-cell + spatial intratumor-heterogeneity atlas, n=7, 1 day fresh) — a methodologically rare longitudinal design for treatment-resistance modeling. Cross-comparison of today's hits.jsonl vs the prior 5-entry window: 4 NOVEL score-9 (GSE343370, GSE343371, GSE308275, GSE343503) + 14 RECYCLED score-9 (GSE330876, GSE311507, GSE328275, GSE337336, GSE334923, GSE306130+GSE316922, GSE292589, GSE328422, GSE277080, GSE281462-GSE281465) + 3 fresh score-3 (GSE343367 JAX scRNAseq male KOLF2.2J hiPSC-derived; GSE293914 CCL20 drug-tolerant persisters EGFR-mutated immunotherapy; GSE331354 transient epithelial plasticity uterus). The full score-9 ladder retained from yesterday's id:67 plus id:65/id:64/id:62 confirms the chronic-IPF, refractory-cancer, and spatial-QC atlases remain the dominant watchlist. Combined four-axis score (peptide / AI-agent infrastructure / longevity / cost): GSE343370+GSE343371 7/12 + GSE308275 6/12 + GSE343503 6/12 = 19/36. NOVEL 3/3 vs the prior 5-entry digest window — passes the goal criterion of ≥2/3 NOVEL. HuggingFace intel tracker (RECYCLED entity / refreshed metrics): longevity-db/mouse-muscle-aging-atlas-snRNAseq (201 downloads, NEW 2026-08-14 ingestion-tag), OATML-Markslab/ProteinGym_v1 (2,048 downloads), tyang816/ProteinGym_v1 (57 downloads), ICML2022/ProteinGym (1,058 downloads, 12 likes), OATML-Markslab/ProteinGym_v0.1 (2,980 downloads, 15 likes), genbio-ai/ProteinGYM-DMS (837 downloads), Jianwen/protein_ligand_cofolding_posebusters (268 downloads), SharkieJones/celldega_Visium-HD_hCRC (11 downloads, NEW entity for BB spatial-QC lane). --- ## 🎯 Highlights ### 1. Signal 1 — Bioinformatics + AI Drug Discovery / MSKCC dual-barcoding pooled CRISPR-KO + scRNA-seq atlas: GSE343370 (watcher score 9, n=3, Homo sapiens, 'mixed' modality, pdat 2026/08/12 = 1 day ago, raw H5AD files available, raw_file_availability 'yes', watcher query 'OpenFold weights') + GSE343371 (watcher score 9, n=24, Homo sapiens, single-cell, pdat 2026/08/12 = 1 day ago, raw H5AD files, raw_file_availability 'yes') form the FRESHEST score-9 dual-barcoding / pooled-CRISPR-KO / scRNA-seq-readout companion atlas in the entire 2026-08-14 scan and a direct AI-DD + perturbation-omics + bioinformatics input for the BB AI-DD lane. The GEO title reports a 'dual barcoding system for pooled analysis of gene knockouts in polyclonal populations' (30KO + 30KO_GEM-X) from MSKCC — a pooled perturbation platform that combines two orthogonal barcoding layers (a perturbation barcode identifying the CRISPR guide + a single-cell barcode identifying the cell of origin) to enable combinatorial-loss-of-function screens at single-cell resolution across polyclonal cell populations. This is a fresh perturbation-omics input for AI-DD target nomination: the GEM-X protocol extends the older 30KO design with enhanced guide-coverage, enabling higher-order combinatorial KO screens (gene-gene interaction mapping, synthetic-lethality discovery, paralog-redundancy testing) at single-cell readout scale. The 1-day freshness, the n=3 (30KO_GEM-X) + n=24 (30KO) design, and the H5AD raw-file panel is a tractable input for AI-DD target nomination (perturbation-effect prediction, gene-gene interaction modeling, target-essentiality classification) and complements the existing GEO pooled-screen reference panel: id:65 GSE341139+GSE341616 (HAND2-BMP5-SMAD1/5/9 hepatic stellate), id:65 GSE330687 (PD-L1 metabolic reprogramming NSCLC), id:62 GSE281464 (TFE3/TFEB translocation RCC Perturb-seq), id:63 GSE307120 (fusion-driven translocation RCC scRNA + ChIP). Four-axis score: peptide 0/3, AI-agent infrastructure 3/3, longevity 1/3, low-cost/ease 3/3 = 7/12. Sources: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343370, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343371. ### 2. Signal 2 — Bioinformatics + AI Drug Discovery / Integrated single-cell + spatial ovarian immune landscape: GSE308275 (watcher score 9, n=6, Homo sapiens, single-cell + spatial, pdat 2026/08/12 = 1 day ago, raw_file_availability 'maybe', watcher query 'multiplex imaging spatial') is the FRESHEST score-9 ovarian-cancer / single-cell + spatial / immune-microenvironment atlas in the entire 2026-08-14 scan and a direct refractory-ovarian-cancer + AI-DD + immune-microenvironment input for the BB bioinformatics + AI-DD lane. The GEO title reports 'Integrated single-cell and spatial transcriptomics of ovarian immune landscape reveal role of...' — a fresh integrated single-cell + spatial atlas of the ovarian-cancer immune microenvironment, providing paired scRNA-seq + spatial transcriptomics at sub-cellular resolution to map the immune geography from tumor core → invasive front → stromal compartment in ovarian cancer. Ovarian cancer is the leading cause of gynecologic-cancer mortality in the US (~20,000 new cases/year, ~13,000 deaths/year, 5-year survival ~50% overall, ~30% in advanced disease) and is dominated by high-grade serous ovarian carcinoma (HGSOC, the most-aggressive histologic subtype, ~70% of cases) with homologous-recombination deficiency (HRD, BRCA1/2 mutation or epigenetic silencing) defining the dominant molecular subtype. The standard-of-care pyramid is primary debulking surgery + platinum-taxane chemotherapy ± bevacizumab (Avastin) + PARP-inhibitor maintenance (olaparib/Lynparza, niraparib/Zejula, rucaparib/Rubraca) for HRD-positive disease. Despite the PARP revolution, ~50% of HRD-positive HGSOC patients still progress (the 'PARP-resistance' phenotype), and the immune-cold tumor microenvironment is the dominant barrier to checkpoint-inhibitor efficacy in ovarian cancer (single-agent PD-1/PD-L1 activity < 10%, combination strategies with chemotherapy + bevacizumab under active investigation). The 1-day freshness, n=6 single-cell + spatial design is a tractable input for AI-DD target nomination (T-cell-exhaustion trajectory, Treg/Th17 balance, myeloid-T-cell cross-talk, ovarian-specific immune checkpoints) and complements the existing BB refractory-cancer reference panel: id:67 GSE328275 (TNBC CD45+ nodal immune atlas), id:67 GSE311507 (HNSCC subtype-dependency), id:65 GSE306130+GSE316922+GSE318638 (NMSC Xenium + scRNA + scTCR), id:64 GSE328422 (lymph-node metastasis immune dynamics). Four-axis score: peptide 0/3, AI-agent infrastructure 3/3, longevity 1/3, low-cost/ease 2/3 = 6/12. Source: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE308275. ### 3. Signal 3 — Bioinformatics + AI Drug Discovery / Longitudinal single-cell + spatial intratumor heterogeneity atlas: GSE343503 (watcher score 9, n=7, Homo sapiens, single-cell + spatial, pdat 2026/08/12 = 1 day ago, raw_file_availability 'maybe', watcher query 'multiplex imaging spatial') is the FRESHEST score-9 longitudinal / single-cell + spatial / intratumor-heterogeneity atlas in the entire 2026-08-14 scan and a direct refractory-cancer + AI-DD + treatment-resistance input for the BB bioinformatics + AI-DD lane. The GEO title reports 'Longitudinal single-cell and spatial transcriptomics reveals intratumor heterogeneity...' — a methodologically rare longitudinal design that combines paired scRNA-seq + spatial transcriptomics at multiple timepoints to track how intratumor heterogeneity evolves over the course of disease progression / treatment. The n=7 design suggests a small-cohort but multi-timepoint design (vs the typical single-timepoint atlas), enabling inference of clonal-architecture evolution, treatment-induced plasticity (epithelial-mesenchymal transition, neuroendocrine transdifferentiation, lineage plasticity), and emergence of resistant subpopulations. Intratumor heterogeneity is the dominant mechanism of treatment failure in solid tumors: clonal diversity enables pre-existing resistant subclones to expand under therapy-induced selection pressure (the 'Darwinian' model of tumor evolution), and spatially-distinct ecological niches (hypoxic core, immune-cold stroma, vascular periphery) further amplify heterogeneity. The 1-day freshness, n=7 multi-timepoint design is a tractable input for AI-DD target nomination (clonal-architecture inference, resistant-subclone identification, niche-specific drug sensitivity, longitudinal-trajectory analysis) and complements the existing BB treatment-resistance reference panel: id:67 GSE311507 (HNSCC subtype-dependency), id:67 GSE328275 (TNBC CD45+ nodal immune atlas), id:65 GSE330687 (PD-L1 metabolic reprogramming NSCLC), id:63 GSE307120 (fusion-driven translocation RCC). Four-axis score: peptide 0/3, AI-agent infrastructure 3/3, longevity 1/3, low-cost/ease 2/3 = 6/12. Source: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343503. ### 4. Tracker note — Biotech Infrastructure / HuggingFace AI-DD + longevity intel scan (RECYCLED entities / refreshed metrics): Today's HuggingFace intel scan returned 8 entries spanning the AI-DD + longevity + protein-fitness landscape. Top ProteinGym / protein-fitness benchmarks (RECYCLED entities, refreshed download counts): OATML-Markslab/ProteinGym_v0.1 (2,980 downloads, 15 likes, arXiv:2205.13760, the original ProteinGym benchmark for protein-fitness prediction), OATML-Markslab/ProteinGym_v1 (2,048 downloads, 8 likes, the v1 update with ~3M mutations across substitution + indel benchmarks), ICML2022/ProteinGym (1,058 downloads, 12 likes), genbio-ai/ProteinGYM-DMS (837 downloads, 1 like), tyang816/ProteinGym_v1 (57 downloads, the ProtSSN dataset variant). AI-DD co-folding benchmarks: Jianwen/protein_ligand_cofolding_posebusters (268 downloads, 0 likes, frozen-data revision of the Autoresearch protein-ligand co-folding task on 62 PoseBusters cases — a fresh input for benchmarking boltz-1 / boltz-2 / AlphaFold3-Multimer on ligand co-folding). NEW entity for the BB longevity lane: longevity-db/mouse-muscle-aging-atlas-snRNAseq (201 downloads, NEW 2026-08-14 ingestion-tag, license cc-by-4.0, the Mouse Skeletal Muscle Aging Atlas sn/scRNA-seq dataset — a fresh single-nucleus + single-cell RNA-seq dataset spanning different age groups for cell-type-specific aging analysis, directly complements the BB longevity + sarcopenia reference panel). NEW entity for the BB spatial-QC lane: SharkieJones/celldega_Visium-HD_hCRC (11 downloads, 0 likes, a 10x Genomics Visium-HD human colorectal-cancer dataset for celldega spatial-QC benchmarking — directly complements yesterday's id:64 GSE277080 spatial-QC metrics atlas). This is a RECYCLED entities / refreshed metrics entry plus 2 NEW entities (longevity-db mouse-muscle-aging-atlas, SharkieJones celldega_Visium-HD_hCRC); it does not count toward the novelty gate but is worth tracking as a continuous signal of the BB AI-DD + biotech-infrastructure lane. Recommended BB move: (a) benchmark Jianwen/protein_ligand_cofolding_posebusters on boltz-1 / boltz-2 once the boltz-2 weights are public; (b) ingest longevity-db/mouse-muscle-aging-atlas-snRNAseq to extend the BB sarcopenia + muscle-aging reference panel; (c) benchmark SharkieJones/celldega_Visium-HD_hCRC against yesterday's id:64 GSE277080 spatial-QC metrics for unified spatial-QC standardization. ### 5. Novelty and action gate — All three actionable signals (GSE343370+GSE343371, GSE308275, GSE343503) are absent from the prior id:62-id:67 window at both accession and study-family level (NOVEL 3/3, passes the goal criterion of ≥2/3 NOVEL). Combined four-axis score: GSE343370+GSE343371 7/12 + GSE308275 6/12 + GSE343503 6/12 = 19/36 — solid mid-range, comparable to id:65 (21/36) but lower than id:67's 24/36 score-9 ladder. The single-cell + spatial study family dominates today's NOVEL hits (3/3 with spatial component, vs id:67's 2/3 with spatial), confirming the trend that BB's most-actionable atlas type is paired scRNA-seq + spatial. The 14 RECYCLED score-9 hits (GSE330876, GSE311507, GSE328275, GSE337336, GSE334923, GSE306130+GSE316922, GSE292589, GSE328422, GSE277080, GSE281462-GSE281465) confirm that yesterday's id:67 'fresh score-9 ladder' is now stable — the priority moves to deep-ingesting the spatial + scRNA-seq matrices for cross-platform AI-DD modeling. Recommended BB move: (a) ingest GSE343370+GSE343371 H5AD raw files to reconstruct the MSKCC dual-barcoding pooled-KO perturbation atlas and benchmark against the existing GEO pooled-screen panel (GSE341139+GSE341616, GSE330687, GSE281464, GSE307120) for BB AI-DD + perturbation-omics briefs; (b) ingest GSE308275 single-cell + spatial matrices to map the ovarian-cancer immune geography and integrate with the TNBC CD45+ nodal atlas (GSE328275) + HNSCC subtype-dependency atlas (GSE311507) for a unified gynecologic + breast + head-and-neck refractory-cancer + AI-DD brief; (c) ingest GSE343503 longitudinal single-cell + spatial matrices to model clonal-architecture evolution and integrate with the NSCLC PD-L1 metabolic-reprogramming atlas (GSE330687) for a unified treatment-resistance + AI-DD brief; (d) ingest longevity-db/mouse-muscle-aging-atlas-snRNAseq and SharkieJones/celldega_Visium-HD_hCRC for BB longevity + spatial-QC lane extension; (e) track ProteinGym + Jianwen protein-ligand co-folding benchmarks for boltz-1 / boltz-2 / AlphaFold3-Multimer benchmark standardization. Do not treat the small n (n=3 30KO_GEM-X, n=24 30KO, n=6 ovarian, n=7 longitudinal) as definitive clinical evidence — these are hypothesis-generation inputs that should be benchmarked against existing reference atlases before commissioning any wet-lab follow-up. --- ## 💼 Next Steps - **[Open GSE343370 in GEO (MSKCC 30KO_GEM-X dual-barcoding pooled KO)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343370)** - **[Open GSE343371 in GEO (MSKCC 30KO dual-barcoding pooled KO scRNA-seq)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343371)** - **[Open GSE308275 in GEO (ovarian immune landscape single-cell + spatial)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE308275)** - **[Open GSE343503 in GEO (longitudinal intratumor heterogeneity single-cell + spatial)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE343503)** - **[Track longevity-db/mouse-muscle-aging-atlas-snRNAseq on HuggingFace](https://huggingface.co/datasets/longevity-db/mouse-muscle-aging-atlas-snRNAseq)** - **[Track Jianwen/protein_ligand_cofolding_posebusters on HuggingFace](https://huggingface.co/datasets/Jianwen/protein_ligand_cofolding_posebusters)** - **[Request AI-DD + refractory-cancer + perturbation-omics brief](https://brownbio.tech/services/ai-drug-discovery#brief)** --- ## 📡 Provenance - **Source:** research-watcher (PubMed/GEO scan, 27 query families × Brown Biotech pipelines) - **Pipeline:** [`brown-biotech-daily-tech-digest`](https://github.com/ohbryt/brown-biotech-platform) — 06:00 KST cron - **Generated:** 2026-08-14 06:09 KST - **Repo:** `ohbryt/brown-biotech-platform` --- _Auto-published by `brown_biotech_research_digest_publisher.py` · [brownbio.tech](https://brownbio.tech) · Decision-ready research, daily._
Brown Biotech Research Digest — 2026-08-14
PubMed/GEO scan · research-watcher · 06:00 KST