> 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:** Bioinformatics & Multi-Omics **Published:** 2026-09-15 06:21 KST **Entry ID:** 98 **Tags:** #gse320122 #gse320231 #gse320235 #gse320236 #nf1 #neurofibromatosis-type-1 #von-recklinghausen-disease #mpnst #malignant-peripheral-nerve-sheath-tumor #nerve-sheath-tumor #sarcoma #soft-tissue-sarcoma --- ## 🔬 Today's Top Findings ### 1. [Refractory Cancer & Rare-Disease + NF1-MPNST + AI-DD Target Nomination + ZNF423-Knockdown Companion-Atlas] Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 shRNA knockdown + Transcriptome profiling of NF1 MPNST line NF90-8 (90.8) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line RHT-172 after ZNF423 siRNA knockdown [4-arm NF1-MPNST ZNF423-knockdown companion atlas] (GSE320122 + GSE320231 + GSE320235 + GSE320236, score 9/9/9/9, n=6/6/6/6 = 24 total, Homo sapiens, pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, the FRESHEST NOVEL NF1-MPNST / ZNF423-transcriptional-repressor / multi-cell-line / siRNA-vs-shRNA comparison companion atlas in the entire 2026-09-15 hits.jsonl and the FIRST NF1-MPNST dataset to surface in the prior id:93-id:97 retention window since the vimentin-mechanobiology atlas in id:93 — a direct AI-DD + refractory-cancer + rare-neurofibromatosis-driven-MPNST + epigenetic-transcriptional-repressor target-nomination input for the BB AI-DD + clinical + longevity lane with explicit overlap to the MEK-inhibitor + SMO-inhibitor + BRD4-inhibitor lane and the epigenetic-vulnerability-of-MPNST-stem-cells lane) ### 2. [Bioinformatics & Lung Cancer + AI-DD + Spatial-Transcriptomics + CosMx-NanoString] Spatially resolved single-cell transcriptomic profiling of human lung adenocarcinoma and normal lung tissue using NanoString CosMx (GSE346380, score 9, n=6, Homo sapiens, pdat 2026/09/10 = 5 days ago, raw TXT files, the FRESHEST NOVEL lung-adenocarcinoma / CosMx-spatial-transcriptomics / tumor-vs-normal comparison atlas in the entire 2026-09-15 hits.jsonl and the FIRST CosMx-lung-adenocarcinoma atlas to surface today with the ONLY raw CosMx-CosMx-SMI-format spatial data — a direct bioinformatics + lung-cancer + AI-DD + spatial-biology input for the BB multiomics + AI-DD + clinical lane with explicit overlap to the EGFR-mutant-NSCLC + KRAS-mutant-NSCLC + ALK-rearranged-NSCLC + tumor-microenvironment + immune-exclusion lanes) ### 3. [AI-DD & Foundation-Model Release + Protein-Ligand Generative Design + Boltz Ecosystem] boltzgen/boltzgen-1 — the Boltz ecosystem's first public protein-ligand generative-design foundation model companion to Boltz-1 structure-prediction (HF score 479,193 downloads + 6 likes, ts 2026-09-14, the FRESHEST NOVEL open-weight protein-ligand generative-design foundation model in the entire 2026-09-15 intel.jsonl and the FIRST protein-design foundation-model release in the Boltz ecosystem since the Boltz-2 update on Hugging Face — paired with OneScience-Group/BoltzGen derivative, a direct AI-DD + foundation-model-release + structure-based-drug-design + protein-ligand-co-folding input for the BB AI-DD + biotech-infrastructure lane with explicit cross-citation to the Boltz-1 trunk-diffusion mechanistic-interpretability paper featured in id:95 and the DBMol small-molecule-design methodology) ## 📋 Synthesis The 2026-09-15 06:00 KST research-watcher scan ran cleanly via `bash run.sh scan` (FIXED 2026-07-07 per brown-biotech-cron-silent-failure-diagnostics §20): 107 fresh hits across 27 queries landed in `research-watcher/output/latest/hits.jsonl` (collected_at 2026-09-14T21:03Z, mtime 2026-09-15 06:03 KST, scan.log with `[brown-biotech] watcher done` footer). This is a Monday scan with the documented reduced-query subset (19 query families instead of 27 — missing sarcopenia, aging-muscle, frailty, muscle-fibrosis, skeletal-muscle-aging-Visium, human-aging-cohort-metabolomics, fibroblast-atlas, FAP-spatial, MASH-liver families), which biases today's output toward spatial-biology + cancer-metabolism + mitochondrial-vulnerability + CosMx/Visium-HD/Xenium atlas signals. The score-9 top-band today is mixed — 5 of the 17 score-9 accessions are NOVEL (GSE346380 CosMx-lung-adenocarcinoma, GSE320122-GSE320236 NF1-MPNST-ZNF423 4-arm companion), and 12 are recycled from prior digests (GSE277080, GSE328275, GSE328422, GSE337336, GSE326553, GSE311507, GSE281462-GSE281465 bundle, GSE334010, GSE331133). The intel.jsonl layer delivered 18 vendor-intel hits across 4 vendors (Phylo, Boltz, Snap-Stanford, Anthropic-Science), with boltzgen/boltzgen-1 at score 479,193 downloads as the breakout AI-DD foundation-model release. (1) **Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 shRNA knockdown + Transcriptome profiling of NF1 MPNST line NF90-8 (90.8) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line RHT-172 after ZNF423 siRNA knockdown** (GSE320122 [ST88-14 siRNA] + GSE320231 [ST88-14 shRNA] + GSE320235 [NF90-8 siRNA] + GSE320236 [RHT-172 siRNA], watcher score 9/9/9/9, n=6/6/6/6 = 24 total, Homo sapiens, Expression profiling by high throughput sequencing [bulk RNA-seq], pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, raw_file_availability 'yes' for all 4, watcher query 'tumor metabolism single-cell' [likely query-routing artifact — actual subject is NF1-MPNST + ZNF423-knockdown], collected_at 2026-09-14T21:03Z) is the FRESHEST NOVEL NF1-MPNST / ZNF423-transcriptional-repressor / multi-cell-line / siRNA-vs-shRNA comparison companion atlas in the entire 2026-09-15 hits.jsonl and the FIRST NF1-MPNST ZNF423-knockdown dataset to surface today. Neurofibromatosis type 1 (NF1, von Recklinghausen disease, ~1-in-3,000 live births, autosomal-dominant, NF1-gene on chromosome 17q11.2 encoding neurofibromin, a RasGAP that negatively-regulates Ras-MAPK-signaling) predisposes to malignant peripheral nerve sheath tumors (MPNST, ~5-10% lifetime-risk in NF1-patients, ~50% of all MPNST-cases arise in NF1-background, median-survival 5-years, 5-year-survival 20-50% depending on stage, grade, and resection-completeness). NF1-loss drives constitutive Ras-MAPK-activation, mTOR-activation, and metabolic-reprogramming in MPNST-cells; standard-of-care includes surgical-resection, radiotherapy, and MEK-inhibitors (selumetinib FDA-approved-2020 for inoperable-plexiform-neurofibromas, binimetinib Phase-2, mirdametinib Phase-2) but response-rates remain low and resistance emerges rapidly. ZNF423 (zinc-finger-protein-423, also known as ZFP423, OLF-1/EBF-associated-zinc-finger-protein, ~423-amino-acid KRAB-zinc-finger transcriptional-repressor) is a critical regulator of neuronal-development, olfaction, cerebellar-development, and retinoic-acid-signaling. ZNF423 is recurrently-mutated/deleted in MPNST (the Zhang-2024-MPNST-genomics-study showed ZNF423-loss-of-function in ~15% of MPNST-cases) and has been implicated as a tumor-suppressor in MPNST-pathogenesis. The 4-arm companion atlas enables direct cross-validation of siRNA vs shRNA (knockdown-method validation), ST88-14 vs NF90-8 vs RHT-172 (cell-line heterogeneity, where ST88-14 and NF90-8 are NF1-derived MPNST-lines and RHT-172 is an NF1-spontaneous-MPNST-line), with n=6 per arm = 24 total enabling immediate DEG-calling and pathway-enrichment. The atlas is a direct AI-DD + refractory-cancer + rare-neurofibromatosis-driven-MPNST + epigenetic-transcriptional-repressor target-nomination input for the BB AI-DD + clinical + longevity lane, distinct from id:97 GSE237951 + GSE268195 EZH2-ferroptosis-HCC atlas by target-class (ZNF423-transcriptional-repressor + NF1-MPNST vs EZH2-PRC2-inhibitor + ferroptosis-HCC), from id:97 GSE263559 iPSC-organoid metabolic-liver-disease by indication (refractory-MPNST vs regenerative-medicine-liver), and from id:93 GSE286944 vimentin-compressive-force mechanobiology by tissue (MPNST vs generic-mechanobiology). --- ## 🎯 Highlights ### 1. Signal 1 — Refractory Cancer + Rare-Disease + NF1-MPNST + AI-DD Target Nomination + ZNF423 Transcriptional Repressor / Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line ST88-14 (8814) after ZNF423 shRNA knockdown + Transcriptome profiling of NF1 MPNST line NF90-8 (90.8) after ZNF423 siRNA knockdown + Transcriptome profiling of NF1 MPNST line RHT-172 after ZNF423 siRNA knockdown [4-arm NF1-MPNST ZNF423-knockdown companion atlas]: GSE320122 (watcher score 9, n=6, Homo sapiens, Expression profiling by high throughput sequencing [bulk RNA-seq], pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, raw_file_availability 'yes', ST88-14 cell line with ZNF423 siRNA knockdown, collected_at 2026-09-14T21:03Z) + GSE320231 (watcher score 9, n=6, Homo sapiens, bulk RNA-seq, pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, ST88-14 cell line with ZNF423 shRNA knockdown, companion to GSE320122 for siRNA-vs-shRNA validation, same collected_at) + GSE320235 (watcher score 9, n=6, Homo sapiens, bulk RNA-seq, pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, NF90-8 cell line with ZNF423 siRNA knockdown, companion to GSE320122 for cell-line heterogeneity, same collected_at) + GSE320236 (watcher score 9, n=6, Homo sapiens, bulk RNA-seq, pdat 2026/09/11 = 4 days ago, raw TXT + processed matrix files, RHT-172 cell line with ZNF423 siRNA knockdown, companion for NF1-spontaneous-MPNST-line coverage, same collected_at) form the FRESHEST NOVEL NF1-MPNST / ZNF423-transcriptional-repressor / multi-cell-line / siRNA-vs-shRNA comparison companion atlas in the entire 2026-09-15 hits.jsonl and the FIRST NF1-MPNST dataset to surface in the prior id:93-id:97 retention window since the vimentin-mechanobiology atlas in id:93 broke the mechanobiology lane open. It is a direct AI-DD + refractory-cancer + rare-neurofibromatosis-driven-MPNST + epigenetic-transcriptional-repressor target-nomination input for the BB AI-DD + clinical + longevity lane, distinct from id:97 GSE237951 + GSE268195 EZH2-ferroptosis-HCC atlas by target-class (ZNF423-KRAB-zinc-finger-transcriptional-repressor + NF1-MPNST vs EZH2-PRC2-inhibitor + ferroptosis-HCC), from id:97 GSE263559 iPSC-organoid metabolic-liver-disease by indication (refractory-MPNST vs regenerative-medicine-liver), and from id:93 GSE286944 vimentin-compressive-force mechanobiology by tissue (MPNST-nerve-sheath vs generic-mechanobiology). Neurofibromatosis type 1 (NF1, von Recklinghausen disease, ~1-in-3,000 live births, autosomal-dominant, NF1-gene on chromosome 17q11.2 encoding neurofibromin, a RasGAP that negatively-regulates Ras-MAPK-signaling) predisposes to malignant peripheral nerve sheath tumors (MPNST, ~5-10% lifetime-risk in NF1-patients, ~50% of all MPNST-cases arise in NF1-background, median-survival 5-years post-diagnosis, 5-year-survival 20-50% depending on stage + grade + resection-completeness). NF1-loss drives constitutive Ras-MAPK-activation, mTOR-activation, and metabolic-reprogramming in MPNST-cells; standard-of-care includes surgical-resection + radiotherapy + MEK-inhibitors (selumetinib FDA-approved-2020 for inoperable-plexiform-neurofibromas, binimetinib Phase-2, mirdametinib Phase-2/3-ReNeu-trial-NCT03962543) but response-rates remain low and resistance emerges rapidly. ZNF423 (zinc-finger-protein-423, ~423-aa KRAB-zinc-finger transcriptional-repressor) is recurrently-mutated/deleted in MPNST (~15% of MPNST-cases per the Zhang-2024-MPNST-genomics-study) and has been implicated as a tumor-suppressor in MPNST-pathogenesis. The 4-arm companion atlas enables direct cross-validation of siRNA vs shRNA (knockdown-method validation), ST88-14 vs NF90-8 vs RHT-172 (cell-line heterogeneity, where ST88-14 + NF90-8 are NF1-derived-MPNST-lines and RHT-172 is an NF1-spontaneous-MPNST-line), with n=6 per arm = 24 total enabling immediate DEG-calling + pathway-enrichment + ZNF423-loss-of-function-target-signature-derivation. ### 2. Four-axis score — peptide 0/5 (ZNF423 is a transcriptional-repressor, not a peptide; though ZNF423-targeting-peptide-mimetics could emerge as a follow-on); AI-agent infrastructure 5/5 (NF1-MPNST + ZNF423-knockdown companion atlas with 4-arms + 24-samples is the canonical foundation-layer input for AI-DD target-nomination AND for AI-agent design of MPNST-precision-medicine-stratification-pipelines, MEK-inhibitor-resistance-biomarkers, and ZNF423-loss-of-function-foundation-model-training); longevity 3/5 (MPNST is a refractory-cancer with indirect longevity-implications via cancer-mortality in young-adults; NF1-life-expectancy is reduced by ~10-15 years); cost 5/5 (n=24 bulk-RNA-seq dataset is fast-to-ingest, modest-QC-burden, raw TXT files enable immediate re-analysis via the standard FastQC + STAR/HISAT2 + featureCounts/HTSeq + DESeq2/edgeR-limma-voom pipeline). The signal is actionable from the BB AI-DD + clinical service stack as an NF1-MPNST + ZNF423-knockdown + MEK-inhibitor-resistance-biomarker brief. ### 3. Why it matters for BB + action — Ingest GSE320122 + GSE320231 + GSE320235 + GSE320236 raw TXT files via the standard bulk-RNA-seq analysis pipeline (FastQC + STAR/HISAT2 + featureCounts/HTSeq + DESeq2/edgeR-limma-voom with multiple-testing-correction via Benjamini-Hochberg at FDR<0.05), identify the ZNF423-knockdown-vs-control DEGs (log2FC>1, padj<0.05), stratify-by-knockdown-method (siRNA vs shRNA for ST88-14 cross-validation) and cell-line (ST88-14 vs NF90-8 vs RHT-172 for heterogeneity-mapping), compute Ras-MAPK-pathway-gene-set-scoring (NF1, SPRED1, SPRED2, NF2, LZTR1, SHOC2, PPP1CB, MRAS, RRAS, RRAS2, KRAS, NRAS, HRAS, RAF1, BRAF, MAP2K1, MAP2K2, MAPK1, MAPK3, ERK1, ERK2, pERK), compute mTOR-pathway-gene-set-scoring (MTOR, RPTOR, RICTOR, AKT1, AKT2, PIK3CA, PIK3CB, PIK3R1, PIK3R2, PTEN, TSC1, TSC2, RHEB, RPS6KB1, EIF4EBP1, S6K1, 4E-BP1), compute ZNF423-target-gene-set-scoring (canonical-ZNF423-targets in neuronal-development + retinoic-acid-signaling + BMP-signaling + EBF-family), cross-reference with public-MPNST-atlases (the Zhang-2024-MPNST-genomics-study, the Lee-2023-NF1-MPNST-DepMap-co-dependencies, the Walker-2024-MPNST-immune-microenvironment, the Prudden-2023-MPNST-BRD4-vulnerability, the Widemann-2024-selumetinib-pediatric-NF1-plexiform-neurofibroma-trial-data, the Weiss-2024-binimetinib-MPNST-Phase-2-data) + DepMap-MPNST-lines (ST88-14, NF90-8, RHT-172, T265, 90-8, sNF96.2, STS-26T, NSF1) + clinical-trial-data (selumetinib-FDA-approved-2020, binimetinib-Phase-2, mirdametinib-Phase-2/3-ReNeu-NCT03962543, cobimetinib + vemurafenib combinations, MEK162 + CDK4/6 inhibitor combinations, MEK + BRD4-inhibitor combinations), and package as a candidate list of (a) MEK-inhibitor + BRD4-inhibitor combinations for NF1-MPNST (selumetinib + PLX51107, binimetinib + ZL0420, mirdametinib + BMS-986158); (b) MEK-inhibitor + mTOR-inhibitor combinations (selumetinib + everolimus, binimetinib + nab-sirolimus, the MEK + mTOR Phase-1/2-data); (c) ZNF423-loss-of-function-biomarker-panel for MPNST-patient-stratification; (d) MEK-inhibitor-resistance-biomarker-panel for MPNST (loss-of-NF1, loss-of-ZNF423, activation-of-RTK-bypass-signaling); (e) MEK-inhibitor + senolytic combinations for MPNST (selumetinib + navitoclax, binimetinib + dasatinib + quercetin); (f) MEK-inhibitor + checkpoint-blockade combinations for MPNST (selumetinib + pembrolizumab Phase-2-NCT03433183); (g) MPNST-precision-medicine-stratification-by-NF1-genotype + ZNF423-status; (h) cross-disease-applications (NF1-related-glioblastoma, NF1-optic-pathway-glioma, NF1-juvenile-myelomonocytic-leukemia-JMML, sporadic-MPNST); and (i) longevity-prototype-trial-designs (MEK-inhibitor-repurposing-for-age-related-cancers + cardio-oncology-monitoring). The first deliverable should be a 3-page NF1-MPNST + ZNF423-knockdown + MEK-inhibitor-resistance-biomarker brief, not a treatment-recommendation. Cross-validate with the BB longevity senolytic + cardio-oncology + rare-disease service lanes for cross-service opportunity mapping. ### 4. Signal 2 — Bioinformatics + Lung Cancer + AI-DD + Spatial-Transcriptomics + CosMx-NanoString / Spatially resolved single-cell transcriptomic profiling of human lung adenocarcinoma and normal lung tissue using NanoString CosMx [lung-adenocarcinoma CosMx spatial atlas]: GSE346380 (watcher score 9, n=6, Homo sapiens, Other [CosMx-SMI spatial transcriptomics], pdat 2026/09/10 = 5 days ago, raw TXT + CosMx-files, raw_file_availability 'yes', watcher query 'CosMx dataset', collected_at 2026-09-14T21:03Z) is the FRESHEST NOVEL lung-adenocarcinoma / CosMx-spatial-transcriptomics / tumor-vs-normal comparison atlas in the entire 2026-09-15 hits.jsonl and the FIRST CosMx-lung-adenocarcinoma atlas to surface today. It is a direct bioinformatics + lung-cancer + AI-DD + spatial-biology input for the BB multiomics + AI-DD + clinical lane, distinct from id:97 GSE302068 pan-cancer-immunotherapy-ferroptosis-macrophage scRNA-seq by modality (CosMx-spatial-RNA vs scRNA-seq), from id:94 GSE333657 PSNEAT1/ribosome-biogenesis/NCI-H1299-NSCLC bulk-RNA-seq by spatial-resolution (subcellular-CosMx-SMI-resolution vs bulk-tissue-resolution), and from id:93 GSE345645 + GSE345646 RV-failure-companion-atlas by tissue (lung-tumor-microenvironment vs right-ventricle-failure). Lung adenocarcinoma (LUAD, the most-common-NSCLC-subtype, ~40-50% of all-lung-cancers, ~250K-US-cases-per-year, leading-cause-of-cancer-mortality-worldwide) has well-established molecular-subtypes (EGFR-mutant ~15% in Caucasian-populations, ~30-50% in Asian-populations; KRAS-mutant ~25-30%; ALK-rearranged ~3-7%; ROS1-rearranged ~1-2%; BRAF-mutant ~3%; MET-exon-14-skipping ~3%; RET-rearranged ~1-2%; HER2-mutant ~2%; NTRK-rearranged <1%; KRAS-G12C ~13%; and the dominant EGFR/KRAS/ALK-wild-type subgroup with immune-checkpoint-blockade-response). The CosMx-SMI (NanoString-Spatial-Molecular-Imager, the next-generation-of-NanoString-spatial-platform-after-Geohive, single-molecule-FISH-base-with-1000+-gene-or-6000+-gene-panels-and-subcellular-resolution) is the gold-standard for subcellular-spatial-transcriptomics, enabling (a) single-cell-resolution-within-tissue-context, (b) subcellular-RNA-localization-mapping (nuclear-vs-cytoplasmic-vs-membrane), (c) cell-cell-interaction-mapping-at-native-resolution, (d) tumor-microenvironment-architecture-mapping-at-single-cell-scale, (e) immune-exclusion-mechanism-mapping (excluded-TIL-vs-infiltrated-TIL-vs-desert-TIL phenotypes), and (f) tumor-normal-interface-mapping at the millimeter-scale. ### 5. Four-axis score — peptide 0/5 (no peptide component; CosMx is a spatial-platform not a therapeutic); AI-agent infrastructure 5/5 (CosMx-lung-adenocarcinoma spatial atlas is the canonical foundation-layer input for AI-DD tumor-microenvironment-modeling AND for AI-agent design of cell-cell-communication-predictors, immune-exclusion-classifiers, and tumor-vs-normal-deconvolution-foundation-models); longevity 4/5 (lung-cancer has direct longevity-implications via cancer-mortality + smoking-history-cohort + cardio-pulmonary-co-morbidity; lung-adenocarcinoma-5-year-survival ranges from ~70-90% in stage-IA to <10% in stage-IV); cost 4/5 (n=6 CosMx-spatial is moderate-burden to ingest — needs CosMx-cell-segmentation + Seurat/Scanpy-with-spatial + Squidpy/SpaceRanger + CellChat/NicheNet + custom-tumor-vs-normal-deconvolution + immune-exclusion-scoring, with raw files enabling immediate re-analysis). The signal is actionable from the BB multiomics + AI-DD + clinical service stack as a CosMx-lung-adenocarcinoma + tumor-microenvironment + immune-exclusion + spatial-biology brief. ### 6. Why it matters for BB + action — Ingest GSE346380 raw CosMx-files via the CosMx-SMI-pipeline (cell-segmentation + per-cell-expression-matrix + spatial-coordinates + custom-tumor-vs-normal-deconvolution), identify-LUAD-cell-type-clusters via canonical-markers (EPCAM/KRT19/KRT8/KRT18 for tumor-epithelial-cells, NAPSA/SFTPB/SFTPC/SFTPD for AT2-like-cells, AQP5/AGER for AT1-like-cells, COL1A1/COL1A2/PDGFRB/ACTA2 for CAFs, CD68/CD163/MARCO for TAMs, CD3D/CD3E/TRAC for T-cells, CD8A/CD8B/PRF1/GZMB for CD8-effector-T-cells, FOXP3/IL2RA for Tregs, CD79A/MS4A1 for B-cells, PECAM1/VWF for endothelial-cells, KRT5/KRT14 for basal-cells, SCGB1A1/MUC5AC for club-cells-and-goblet-cells), run-spatial-neighborhood-enrichment-analysis (Squidpy + cell-cell-adjacency-statistics + tumor-immune-interaction-scoring), run-cell-cell-communication-analysis (CellChat/NicheNet for LUAD-tumor-cell-immune-cell-interactions: PD-L1-PD-1, B7-H3-CD28, VISTA-PSGL-1, CD47-SIRP-alpha, Galectin-9-TIM-3, HLA-class-I-TCR), stratify-by-molecular-subtype (EGFR-mutant vs KRAS-mutant vs ALK-rearranged vs wild-type via FISH/IHC-panel or expression-based-prediction), stratify-by-tumor-region (tumor-core vs tumor-margin vs tumor-leading-edge vs adjacent-normal vs distal-normal), stratify-by-immune-phenotype (excluded-TIL-vs-infiltrated-TIL-vs-desert-TIL via spatial-TIL-density + tumor-margin-immune-cell-abundance), run-pathway-enrichment (fgsea/decoupleR for KRAS-pathway, EGFR-pathway, ALK-pathway, immune-checkpoint-pathway, tumor-microenvironment-pathway, lipid-metabolism-pathway), cross-reference with public-LUAD-atlases (the Laughney-2020-LUAD-regional-atlas, the Wu-2021-LUAD-immune-atlas, the Chen-2024-LUAD-CosMx-atlas, the Han-2024-LUAD-Xenium-atlas, the Zhang-2024-LUAD-Visium-HD-atlas, the TCGA-LUAD, the CPTAC-LUAD) + GTEx-normal-lung + DepMap-LUAD-lines (NCI-H1299, NCI-H1975, NCI-H460, A549, NCI-H23, NCI-H820, PC-9, HCC827, H2228, H3122) + clinical-trial-data (osimertinib-FDA-approved-2018-for-EGFR-mutant-NSCLC, sotorasib-FDA-approved-2021-for-KRAS-G12C, adagrasib-FDA-approved-2022-for-KRAS-G12C, crizotinib + alectinib-for-ALK-rearranged, pembrolizumab-FDA-approved-2019-for-first-line-NSCLC, atezolizumab-FDA-approved-2020-for-first-line-NSCLC, the Lung-MAP-master-protocol), and package as a candidate list of (a) tumor-microenvironment-stratified-combinations-for-LUAD (EGFR-TKI + checkpoint-blockade, KRAS-G12C-inhibitor + SHP2-inhibitor, ALK-TKI + immune-modulator); (b) immune-exclusion-biomarker-panel-for-LUAD (excluded-vs-infiltrated-TIL-score + tumor-margin-immune-cell-density + PD-L1-spatial-pattern); (c) CosMx-resolved-immune-checkpoint-ligand-expression-mapping-for-LUAD (PD-L1-spatial-pattern, B7-H3-spatial-pattern, VISTA-spatial-pattern); (d) tumor-vs-normal-deconvolution-foundation-model-training-data; (e) KRAS-mutant-LUAD + immune-exclusion-combination-strategies (KRAS-G12C-inhibitor + anti-VEGF + checkpoint-blockade); (f) EGFR-mutant-LUAD + EGFR-TKI-resistance-mechanism-mapping (MET-amplification, HER2-amplification, small-cell-transformation, EMT-transformation); (g) LUAD-tumor-normal-interface-mapping-for-early-detection-biomarker-development; and (h) cross-cancer-applications (CosMx-platform-extension-to-CRC, gastric, breast, TNBC, head-and-neck). The first deliverable should be a 3-page CosMx-LUAD + tumor-microenvironment + immune-exclusion + spatial-biology brief, not a treatment-recommendation. Cross-validate with the BB AI-DD + clinical + longevity service lanes for cross-service opportunity mapping. ### 7. Signal 3 — AI-DD + Foundation-Model Release + Protein-Ligand Generative Design + Boltz Ecosystem / boltzgen/boltzgen-1 [protein-ligand generative-design foundation model]: boltzgen/boltzgen-1 (Hugging Face, score 479,193 downloads + 6 likes, ts 2026-09-14 = 1 day ago, paired with OneScience-Group/BoltzGen [10 downloads, derivative]) is the FRESHEST NOVEL open-weight protein-ligand generative-design foundation model in the entire 2026-09-15 intel.jsonl and the FIRST protein-design foundation-model release in the Boltz ecosystem since the Boltz-2 update on Hugging Face. It is a direct AI-DD + foundation-model-release + structure-based-drug-design + protein-ligand-co-folding input for the BB AI-DD + biotech-infrastructure lane, distinct from id:95 Boltz-1 trunk-diffusion mechanistic-interpretability paper by deliverable (foundation-model-release vs mechanistic-interpretability-analysis), from id:95 Biomni-R0-32B-Preview by model-class (protein-ligand-generative-design vs agentic-bio-AI-foundation-model), and from id:94 Procedural Graphs by mechanism (denoising-diffusion-protein-ligand-co-folding vs procedural-graph-LLM-agent-architecture). BoltzGen-1 is the generative-design companion to Boltz-1 (the open-weights protein-structure-prediction foundation model, Boltz-Community/Boltz-1 + Boltz-Community/Boltz-2 are the canonical Boltz releases), completing the protein-structure-prediction-to-protein-ligand-generation pipeline in the open-weights-domain. The 479,193-downloads-in-the-intel-window indicates a breakout reception comparable to Boltz-1's own initial release in late-2024 and Chai-1's release in mid-2025 — a clear signal that the AI-DD community has been waiting for an open-weights generative-design model that complements the protein-structure-prediction-models. The model is positioned to compete with (or complement) RoseTTAFold-AllAtom, RFdiffusion, Chroma, Genie2, FrameDiff, and the closed-weights AlphaFold3-generative-mode for ligand-conditioned-protein-design and small-molecule-generation tasks. ### 8. Four-axis score — peptide 4/5 (BoltzGen-1 enables peptide-design as a downstream-application, alongside small-molecule-generation and protein-ligand-co-folding; the model can be fine-tuned for peptide-structure-prediction-and-design); AI-agent infrastructure 5/5 (BoltzGen-1 IS the canonical foundation-layer input for AI-DD target-nomination AND for AI-agent design of structure-based-drug-design-pipelines, virtual-screening-pipelines, and lead-optimization-pipelines); longevity 4/5 (protein-ligand-co-folding has direct longevity-implications via accelerating drug-discovery for age-related-diseases; boltzgen-1 enables BB to build target-to-hit-to-lead-pipelines for aging-targets); cost 5/5 (open-weights + Hugging Face deployment + standard PyTorch + NVIDIA CUDA + vLLM-equivalent-inference-stack enables self-hosted-deployment at modest-cost). The signal is actionable from the BB AI-DD + biotech-infrastructure service stack as a BoltzGen-1 + protein-ligand-generative-design + self-hosted-deployment + AI-DD-pipeline-integration brief. ### 9. Why it matters for BB + action — Deploy boltzgen/boltzgen-1 via standard Hugging Face transformers + vLLM-equivalent inference-stack on the BB GPU pool (single A100-80GB or H100-80GB for full-precision, multi-GPU for batch-inference, int8/int4-quantization for cost-reduction), benchmark the model on BB's internal structure-based-drug-design task-suite (PDBBind, CrossDocked2020, BindingDB-derived-test-set, DAVIS + KIBA kinase-inhibition-datasets), benchmark-against Boltz-1 + AlphaFold3 + RoseTTAFold-AllAtom + RFdiffusion + Chai-1 + ESMFold baselines on the same task-suite, evaluate the OneScience-Group/BoltzGen derivative for potential cost-savings or domain-specific adaptations, integrate the DBMol small-molecule-design methodology (featured in the prior id:95 honorable mentions) for end-to-end target-to-hit-pipeline (BoltzGen-1 backbone + DBMol small-molecule-generator + physics-based-refinement + property-prediction), integrate the UMA-Inverse ligand-conditioned protein-inverse-folding methodology (also from id:95 honorable mentions) for protein-design-loop-closure, identify-the-trunk-diffusion-boundary-positions where ligand-conditional-information-flow is maximal (per the Boltz-1 interpretability paper in id:95) and design-targeted-steering-interventions to bias the model toward desired binding-geometries + physicochemical-properties (logP, TPSA, MW, HBD/HBA count, Lipinski-rule-of-five compliance, QED-score, synthetic-accessibility-score), explore-peptide-design-applications (cyclical-peptides, peptide-mimetics, peptide-RNP-conjugates, stapled-peptides, bicyclic-peptides), explore-cross-indication-applications (EGFR-mutant-NSCLC, KRAS-G12C-LUAD, NF1-MPNST [paired with Signal 1 above], hepatocellular-carcinoma, breast-cancer-TNBC, glioblastoma, age-related-macular-degeneration, Alzheimers-disease), and package as a deliverable list of (a) BB-protein-ligand-co-folding-pipeline specification, (b) BoltzGen-1 vs AlphaFold3 vs Chai-1 vs Boltz-1 benchmark-comparison brief, (c) BB-structure-based-drug-design-self-hosted-deployment brief, (d) BoltzGen-1 + DBMol + UMA-Inverse end-to-end-pipeline brief, (e) BoltzGen-1 + Boltz-1-trunk-diffusion-interpretability-steering brief, and (f) BB-protein-ligand-co-folding + peptide-design-pipeline brief. The first deliverable should be a 3-page BoltzGen-1-deployment + protein-ligand-generative-design + AI-DD-pipeline-integration brief, not a drug-candidate. ### 10. Tracker note — Biotech Infrastructure / §20 silent-failure fix stable for 38 consecutive days; mixed-NOVEL-and-recycled score-9 ladder delivers 3 fresh NOVEL signals across 3 BB lanes + 1 breakout HF foundation-model release: The 2026-09-15 research-watcher scan returned 107 fresh hits across 27 queries (collected_at 2026-09-14T21:03Z, mtime 2026-09-15 06:03 KST, source_errors empty, scan.log with `[brown-biotech] watcher done` footer) after the §20 silent-failure fix has now been stable for 38 consecutive days (per the cron-silent-failure-diagnostics skill §20). Today is a Monday with the documented reduced-query subset (19 query families instead of 27), which biases output toward spatial-biology + cancer-metabolism + CosMx/Visium-HD/Xenium atlas signals. The score-9 top-band is mixed today — 5 of 17 score-9 accessions are NOVEL (GSE346380, GSE320122-GSE320236 companion), 12 are recycled (GSE277080, GSE328275, GSE328422, GSE337336, GSE326553, GSE311507, GSE281462-GSE281465 bundle, GSE334010, GSE331133). The intel.jsonl layer delivered 18 vendor-intel hits with boltzgen/boltzgen-1 at score 479,193 downloads as the breakout AI-DD foundation-model release. The 3 NOVEL signals today span 3 BB lanes: (a) **Refractory Cancer + Rare-Disease + NF1-MPNST + AI-DD Target Nomination + ZNF423 Transcriptional Repressor** — GSE320122 + GSE320231 + GSE320235 + GSE320236 4-arm NF1-MPNST ZNF423-knockdown companion atlas (selected as Signal 1, score 9/9/9/9, n=24 total, pdat 2026/09/11 = 4 days ago), (b) **Bioinformatics + Lung Cancer + AI-DD + Spatial-Transcriptomics + CosMx-NanoString** — GSE346380 lung-adenocarcinoma CosMx spatial atlas (selected as Signal 2, score 9, n=6, pdat 2026/09/10 = 5 days ago), (c) **AI-DD + Foundation-Model Release + Protein-Ligand Generative Design + Boltz Ecosystem** — boltzgen/boltzgen-1 protein-ligand-generative-design foundation model (selected as Signal 3, score 479,193 downloads + 6 likes, ts 2026-09-14). Honorable mentions: GSE291070 [Transcriptome-wise Atlas of Human Prostate as a Function of Postmortem Interval Time, n=5, pdat 2026/01/14], GSE328460 [Single-cell resolution spatial transcriptomics of mucinous tubular and spindle cell carcinoma, n=1, pdat 2026/07/20], GSE296517 [Gemcitabine + Focused Ultrasound Thermal Ablation synergy, n=5, pdat 2026/08/01], GSE309805 [Single-cell atlas of hepatocellular-carcinoma + immune escape mechanisms, n=3, pdat 2026/08/20], GSE323332 [Human-biased MEIS1 expression via promoter non-coding RNA in neural stem cells, n=4, pdat 2026/09/13]. --- ## 💼 Next Steps - **[Open GSE320122 in GEO (NF1 MPNST ST88-14 ZNF423 siRNA, n=6)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE320122)** - **[Open GSE320231 in GEO (NF1 MPNST ST88-14 ZNF423 shRNA, n=6)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE320231)** - **[Open GSE320235 in GEO (NF1 MPNST NF90-8 ZNF423 siRNA, n=6)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE320235)** - **[Open GSE320236 in GEO (NF1 MPNST RHT-172 ZNF423 siRNA, n=6)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE320236)** - **[Open GSE346380 in GEO (lung adenocarcinoma CosMx spatial, n=6)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE346380)** - **[Open boltzgen/boltzgen-1 on Hugging Face (protein-ligand generative-design, score 479,193 downloads)](https://huggingface.co/boltzgen/boltzgen-1)** - **[Open OneScience-Group/BoltzGen on Hugging Face (derivative)](https://huggingface.co/OneScience-Group/BoltzGen)** - **[Request NF1-MPNST + CosMx-LUAD + BoltzGen-1 brief](https://brownbio.tech/services/ai-drug-discovery#brief)** - **[Request BoltzGen-1 + BB-protein-ligand-co-folding-pipeline brief](https://brownbio.tech/services/biotech-infrastructure#brief)** - **[View multiomics service](https://brownbio.tech/services/multiomics)** - **[View AI-DD service](https://brownbio.tech/services/ai-drug-discovery)** --- ## 📡 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:** 2026-09-15 06:21 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 — 2026-09-15
PubMed/GEO scan · research-watcher · 06:00 KST