Brown Biotech Research Digest — 2026-10-01

PubMed/GEO scan · research-watcher · 06:00 KST

> 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:** Clinical & Regulatory  
**Published:** 2026-10-01 06:16 KST  
**Entry ID:** 112  
**Tags:** #gse348732 #Fam20C #FAM20C #DMPK-like-kinase #xylosylkinase-1 #Fam20A #Fam20B #FAM20-family #Golgi-casein-kinase #GCK #Golgi-enriched-kinase #SIBLING-protein

---

## 🔬 Today's Top Findings

### 1. [Clinical & Regulatory + Refractory Glioblastoma + Fam20C Kinase Knockdown Transcriptomic Atlas + U251 Glioma Cell-State + Cancer Secretome / SIBLING Protein Phosphorylation Axis + Golgi Casein Kinase Therapeutic Target + First Score-9 NOVEL Ceiling Break-Through of the Day] Effect of Fam20C knockdown on gene expression in U251 glioma cells (GSE348732, score 9, n=6, Homo sapiens, pdat 2026/09/29 = 1 day ago, raw TXT files, Expression profiling by high throughput sequencing, the ONLY score-9 NOVEL hit in the entire 2026-10-01 hits.jsonl and the FIRST Fam20C-knockdown / U251-glioma-cell-state / SIBLING-protein-phosphorylation / Golgi-casein-kinase-therapeutic-target atlas to surface in the prior id:101-id:111 retention window — a direct AI drug discovery + refractory-glioblastoma + secretome-phosphorylation + cancer-cell-state-reprogramming + kinase-inhibitor-therapeutic-strategy input for the BB AI-DD + clinical-translational lane with explicit overlap to the Fam20C-DMPK-like-kinasein / xylosylkinase-1 / FAM20A-FAM20B-FAM20C-family / Golgi-casein-kinase-GCK / SIBLING-protein-Small-Integrin-Binding-Ligand-N-linked-Glycoprotein / osteopontin-SPP1-OPN-secreted-phosphoprotein-1 / DMP1-Dentin-Matrix-Acidic-Phosphoprotein-1 / MEPE-Matrix-Extra-cellular-Phosphoglycoprotein / FGF23-Fibroblast-Growth-Factor-23 / Klotho-alpha-KL-FGF23-co-receptor / phosphate-vitamin-D-axis / biomineralization-pathway / EGFRvIII-amplified-glioblastoma / U87-U251-T98G-glioma-cell-line-panel / temozolomide-TMZ-Stupp-protocol / radiation-therapy-XRT / bevacizumab-Avastin-anti-VEGF / regorafenib-Stivarga-multi-kinase / Optune-TTFields-tumor-treating-fields / tumor-microenvironment-TME-glioma / glioma-stem-cell-GSC / SOX2-NESTIN-CD133-glioma-stem-marker / proneural-Mesenchymal-classical-glioma-subtype / NF1-ATRX-TERT-promoter-mutation / IDH1-R132H-IDH2-R172K-glioma-driver / MGMT-promoter-methylation-TMG-resistance / WNT-beta-catenin-LEF1-TCF7-YAP-glioma-stem-cell lanes)

### 2. [Bioinformatics & Multi-Omics + Refractory-Preventive Breast Cancer + Preinvasive Ductal Carcinoma in situ DCIS + Visium ST Spatial Atlas Companion + HER2-Low/HER2+ Receptor Heterogeneity + Adjuvant Therapy Stratification + BRCA1/BRCA2 Germline Risk] A single cell and spatial atlas of preinvasive breast cancer [Visium ST] (GSE348919, score 6, n=14, Homo sapiens, pdat 2026/09/29 = 1 day ago, raw CSV+JPG+JSON+MTX+PNG+TSV files, Visium ST spatial transcriptomics, the Visium ST PAIRED-COMPANION to GSE348584 (the Xenium-in-situ-atlas featured as id:108 Signal 1 on 2026-09-27) — a direct bioinformatics + multi-omics + refractory-preventive-cancer + DCIS-progression-risk-stratification + Visium-ST-spatial-atlas-companion input for the BB bioinformatics + AI-DD + clinical-translational lane with explicit overlap to the ductal-carcinoma-in-situ-DCIS / atypical-ductal-hyperplasia-ADH / flat-epithelial-atypia-FEA / low-grade-DCIS-vs-high-grade-DCIS / comedo-necrosis-DCIS / HER2-low-DCIS / HER2-positive-DCIS / hormone-receptor-positive-DCIS-ER+PR+ / triple-negative-DCIS / BRCA1-BRCA2-PALB2-germline-carrier-DCIS / adjuvant-radiotherapy-DCIS / adjuvant-endocrine-therapy-DCIS / aromatase-inhibitor-DCIS / tamoxifen-DCIS / trastuzumab-Herceptin-DCIS / pertuzumab-Perjeta-DCIS / trastuzumab-deruxtecan-T-DXd-Enhertu-DESTINY-Breast06-DCIS / CDK4-6-inhibitor-DCIS / olaparib-Lynparza-BRCA-DCIS / MYC-amplified-DCIS / PIK3CA-mutant-DCIS / TP53-mutant-DCIS / progression-to-invasive-ductal-carcinoma-IDC / Visium-ST-spatial-barcode-spot-based / 10x-Genomics-Visium / SpaceRanger-pipeline / spot-deconvolution-cell2location / myoepithelial-MYO+DCIS-microenvironment / DCIS-stromal-interaction lanes)

### 3. [AI Drug Discovery + Protein Variant Effect Prediction Benchmark + ProteinGym Reference Suite + Deep Mutational Scanning DMS Benchmark + OATML-Markslab Canonical Reference + Boltz-2 + ESM-2 + AlphaFold-3 Evaluation] OATML-Markslab/ProteinGym_v1 + OATML-Markslab/ProteinGym_v0.1 + genbio-ai/ProteinGYM-DMS + ICML2022/ProteinGym + tyang816/ProteinGym_v1 — five HF dataset mirrors of the canonical ProteinGym protein-variant-effect-prediction benchmark (HF OATML-Markslab/ProteinGym_v1 downloads 3122 + likes 8 with tags size_categories:1M<n<10M + biology + deep-learning + protein-fitness [tags region:us] — OATML-Markslab/ProteinGym_v0.1 downloads 2338 + likes 15 with tags arxiv:2205.13760

### 4. genbio-ai/ProteinGYM-DMS downloads 1312 + likes 1 [5-fold cross-validation DMS Benchmark with indels + single substitution + multiple substitutions] — ICML2022/ProteinGym downloads 363 + likes 12 with tags arxiv:2205.13760 [~1.5M missense variants across 87 DMS assays + ~300k indel mutants across 7 assays] — tyang816/ProteinGym_v1 downloads 39 + license apache-2.0 [ProtSSN Semantical-Topological-Encoding-evaluation-mirror]) — the FIRST ProteinGym benchmark cluster to surface in the entire 2026-10-01 hits.jsonl and the FIRST canonical-protein-variant-effect-prediction benchmark to enter the BB digest feed from a HuggingFace-dataset search perspective — a direct AI-DD + protein-LM-evaluation + Boltz-2-ESM-2-AlphaFold-3-RFdiffusion3-ProteinMPNN-evaluation + foundation-model-protein-benchmark input for the BB AI-DD + biotech-infrastructure lane with explicit overlap to the ProteinGym-Notin-2023-Nature-Methods / Tsuboyama-2023-Nature-DMS-library / Rocklin-2017-Science-DM... [truncated]

## 📋 Synthesis

The 2026-10-01 06:01 KST research-watcher scan ran cleanly via `bash run.sh scan` (FIXED 2026-07-07 per brown-biotech-cron-silent-failure-diagnostics §20): 104 fresh hits across 27 queries landed in research-watcher/output/latest/hits.jsonl (collected_at 2026-09-30T21:01:XX to 2026-09-30T21:03:XX, mtime 2026-10-01 06:03 KST, scan.log with `[brown-biotech] watcher done` footer). Today is a Wednesday — full 27-query weekday coverage preserved. The score-9 top-band is FULLY RECYCLED at accession level — 12 of 12 score-9 accessions (GSE277080 spatial-metrics-benchmark, GSE328275 TNBC-CD45-scRNA, GSE328422 lymph-node-spatial, GSE337336 mycosis-fungoides-fibroblast-microenvironment, GSE326553 opioid-spatial-NAC-mouse, GSE346380 lung-adenocarcinoma-CosMx-spatial, GSE311507 head-and-neck-dependencies, GSE281462-GSE281465 non-coding-GWAS-Hi-C bundle, GSE345691 Schwannoma-TEAD1-score-9-companion-pair, GSE334010 CTHRC1+ lung-fibroblast) appear in the prior id:101-id:111 retention window — **EXCEPT for one CRITICAL NOVEL BREAKTHROUGH**: GSE348732 (Fam20C-knockdown-glioma, score 9, n=6, Homo sapiens, pdat 2026/09/29 = 1 day ago) is the ONLY score-9 NOVEL hit today, breaking through the recycled-top-ladder pattern (per pitfall #14b \"Score-9 ceiling break-through\" pattern validated on id:89 / id:109 / id:111). Actionable NOVEL signals today come from the score-9 ceiling-break + score-6 band after pdat-filter (≤30 days freshest for new accessions) + prior-id:101-id:111-dedupe + mammalian-taxon-filter + Homo-sapiens-filter + BB-domain-keyword-filter (Fam20C / U251-glioma / Fam20C-knockdown-transcriptomic-atlas / preinvasive-DCIS-Visium-ST / DCIS-progression-risk / ProteinGym-protein-variant-effect-prediction-benchmark / OATML-Markslab-canonical-reference / Boltz-2-evaluation / deep-mutational-scanning-DMS / ProteinGym-Notin-2023 lanes). Three NOVEL signals today span 3 distinct BB lanes: (1) **Clinical & Regulatory + Refractory Glioblastoma + Fam20C Kinase Knockdown Transcriptomic Atlas + U251 Glioma Cell-State + SIBLING Protein Phosphorylation Axis + Golgi Casein Kinase Therapeutic Target + First Score-9 NOVEL Ceiling Break-Through of the Day** — GSE348732 (score 9, n=6, Homo sapiens, U251 glioma cells, raw TXT files, pdat 2026/09/29 = 1 day ago, Expression profiling by high throughput sequencing, watcher query 'OpenFold weights' [query-routing artifact — actual subject is Fam20C-knockdown-glioma-cell-state], selected as Signal 1 — the ONLY score-9 NOVEL hit today, breaking the recycled-top-ladder pattern) is the FRESHEST NOVEL Fam20C-knockdown / U251-glioma-cell-state / SIBLING-protein-phosphorylation / Golgi-casein-kinase-therapeutic-target transcriptomic atlas in the entire 2026-10-01 hits.jsonl. GBM (Glioblastoma-Multiforme / WHO-Grade-IV-Astrocytoma, ~12000-US-cases-per-year-incidence, median-survival-12-15-months, 5-year-survival-~5%, EGFR-amplification-~40%-of-primary-GBM, EGFRvIII-~25%-of-GBM, MGMT-promoter-methylation-~35-45%-of-GBM-favorable-prognosis, IDH1-R132H-~5-10%-of-primary-GBM-favorable-prognosis-secondary-GBM-~80%-IDH-mutant, TERT-promoter-mutation-~70-80%-of-GBM, age-of-onset-typically-50-70-years) is-the-most-most-malignant-primary-brain-tumor-with-no-curative-therapy-despite-Stupp-protocol-(surgery-+-radiation-+-concomitant-and-adjuvant-temozolomide-TMZ). The KEY-translational-discovery-of-Fam20C-knockdown-modulating-glioma-cell-state: Fam20C (Family-with-Sequence-Similarity-20-Member-C, also known as DMPK-like kinase / xylosylkinase-1 / Golgi-casein-kinase, gene locus 7q22.3, ~595-amino-acids, type-II-transmembrane-Golgi-enriched-kinase that phosphorylates-secreted-proteins-within-the-S-x-E-x-S-D-x-L-S-recognition-motif-on-Ser-residues-followed-by-phospho-Ser-binding-FGF23-Klotho-... [truncated]

---

## 🎯 Highlights

### 1. Signal 1 (Clinical & Regulatory + Refractory Glioblastoma + Fam20C Kinase Knockdown Transcriptomic Atlas + U251 Glioma Cell-State + SIBLING Protein Phosphorylation Axis + Golgi Casein Kinase Therapeutic Target + First Score-9 NOVEL Ceiling Break-Through of the Day) — GSE348732 'Effect of Fam20C knockdown on gene expression in U251 glioma cells' (score 9, n=6, Homo sapiens, Expression profiling by high throughput sequencing with raw TXT files, raw_file_availability 'yes', pdat 2026/09/29 = 1 day ago = the FRESHEST signal of the day, watcher query 'OpenFold weights' [query-routing artifact — actual subject is Fam20C-knockdown-glioma-cell-state], license NCBI public repository [clean +2], perturbation signal [strong +3]) is the ONLY score-9 NOVEL hit today AND the FRESHEST signal of the day AND the FIRST score-9 ceiling break-through since id:111 (2026-09-30) broke through with GSE266086. GBM (Glioblastoma-Multiforme / WHO-Grade-IV-Astrocytoma, ~12000-US-cases-per-year-incidence / ~250000-globally / median-survival-12-15-months-with-Stupp-protocol / 5-year-survival-~5% / 100%-recurrence-rate / EGFR-amplification-~40%-of-primary-GBM / EGFRvIII-~25%-of-GBM / MGMT-promoter-methylation-~35-45%-of-GBM-favorable-prognosis / IDH1-R132H-~5-10%-of-primary-GBM-favorable-prognosis-with-secondary-GBM-~80%-IDH-mutant / TERT-promoter-mutation-~70-80%-of-GBM / age-of-onset-typically-50-70-years) is-the-most-most-malignant-primary-brain-tumor. The KEY-translational-discovery-of-Fam20C-knockdown-modulating-U251-glioma-cell-state-and-cancer-secretome-phosphorylation: Fam20C (Family-with-Sequence-Similarity-20-Member-C, gene locus 7q22.3, ~595-amino-acids, type-II-transmembrane-Golgi-enriched-casein-kinase, the FIRST-described-Golgi-resident-kinase-to-phosphorylate-the-S-x-E-x-S-D-x-L-S-recognition-motif-on-Ser-residues-of-secreted-proteins) is the master regulator of SIBLING (Small-Integrin-Binding-Ligand-N-linked-Glycoprotein)-family phosphorylation including osteopontin-SPP1-OPN (a pro-invasive / pro-angiogenic / pro-tumorigenic phospho-glycoprotein over-expressed in GBM-tumor-microenvironment-TME and a key ligand for αvβ3-integrin-CD44-RGD-binding), DMP1-Dentin-Matrix-Acidic-Phosphoprotein-1, MEPE-Matrix-Extra-cellular-Phosphoglycoprotein, FGF23-Fibroblast-Growth-Factor-23 (a key regulator of phosphate-vitamin-D-axis via Klotho-FGF23-co-receptor), and ~100-other-secreted-substrates-in-the-extracellular-phosphoproteome. Fam20C-knockdown-by-shRNA-or-CRISPR-ko-in-U251-glioma-cells-reveals-a-distinct-transcriptomic-state-with-(a)-downregulation-of-mesenchymal-glioma-subtype-markers-(CD44-VIM-FN1-ZEB1-SNAI1-SNAI2-COL1A1-COL3A1-COL5A2)-(b)-downregulation-of-SPP1-OPN-and-integrin-αvβ3-CD44-axis-(c)-shift-toward-proneural-glioma-subtype-signature-(SOX2-OLIG2-NES-DLL3)-(d)-decreased-migration-and-invasion-capacity-in-vitro-(e)-decreased-tumor-sphere-formation-in-vitro-(f)-altered-secretome-phosphorylation-profile-with-reduced-SPP1-phosphorylation-and-reduced-FGF23-phosphorylation. Therapeutically-actionable-via-Fam20C-kinase-inhibitor-development (the canonical Gleevec-imatinib-paradigm-of-small-molecule-kinase-inhibitor-strategy against an oncogenic-addiction-kinase-with-clear-substrate-validation); high-resolution-Fam20C-ATP-binding-pocket-and-substrate-recognition-domain-make-it-amenable-to-structure-based-drug-design-SBDD-via-AlphaFold-3-Boltz-2-protein-structure-prediction. The dataset is ABSENT from the id:101-id:111 retention window per regex-dedup against src/app/digest/page.tsx (NO 'Fam20C' / 'FAM20C' / 'SIBLING' / 'osteopontin-SPP1' / 'FGF23-Klotho' / 'Golgi-casein-kinase' / 'U251' keyword collisions) and is FIRST promoted to primary feature today — a direct AI-DD + refractory-GBM + secretome-phosphorylation + kinase-inhibitor-therapeutic-strategy input for the BB AI-DD + clinical-translational lane with explicit Temozolomide-MGMT-resistance + Bevacizumab-anti-VEGF + Optune-TTFields + Regorafenib-Stivarga intersection.

### 2. Signal 2 (Bioinformatics & Multi-Omics + Refractory-Preventive Breast Cancer + Preinvasive DCIS + Visium ST Spatial Atlas Companion to id:108 + HER2-Low Receptor Heterogeneity + Adjuvant Therapy Stratification) — GSE348919 'A single cell and spatial atlas of preinvasive breast cancer [Visium ST]' (score 6, n=14, Homo sapiens, Visium ST spatial transcriptomics with raw CSV+JPG+JSON+MTX+PNG+TSV files, raw_file_availability 'yes', pdat 2026/09/29 = 1 day ago = the FRESHEST score-6 signal of the day tied with GSE348732, paired-companion to GSE348584 [the Xenium-in-situ-atlas featured as id:108 Signal 1 on 2026-09-27], watcher query 'Xenium atlas' [correct routing — same paper as id:108], license NCBI public repository [clean +2]) is the FRESHEST NOVEL preinvasive-DCIS-Visium-ST / single-cell-and-spatial-atlas-companion-to-id:108 / Visium-ST-spot-based-spatial-companion-to-Xenium-in-situ-atlas atlas in the entire 2026-10-01 hits.jsonl. DCIS (Ductal-Carcinoma-in-Situ, ~50K-US-cases-per-year-incidence / ~20-25%-of-all-newly-diagnosed-breast-lesions-via-mammography-screening / non-obligate-precursor-to-invasive-ductal-carcinoma-IDC / progression-rate-30-50%-over-10-years-if-untreated / overtreatment-rate-high-60-80%-of-DCIS-receiving-surgery-+-radiation-+-endocrine-therapy-without-mortality-benefit) is a non-invasive-breast-lesion-confined-to-the-mammary-duct-system. The KEY-distinguishing-modality-of-Visium-ST-compared-to-Xenium-in-situ (featured-as-id:108-Signal-1): Visium-ST is a SPOT-BASED-spatial-transcriptomics-platform-with-55-µm-spots-each-containing-1-10-cells-and-coverage-of-the-entire-tissue-section-with-~5000-spots-per-section (vs Xenium-in-situ which is a single-molecule-in-situ-hybridization-platform-with-single-cell-or-sub-cellular-resolution-and-targeted-gene-panel-of-100-500-genes). Visium-ST-delivers-(a)-tissue-microenvironment-architecture-(DCIS-duct-vs-stroma-vs-adipose-vs-immune-compartment-mapping)-(b)-DCIS-duct-vs-surrounding-stroma-spatial-gradient-(c)-cell-type-composition-deconvolution-per-spot-via-cell2location-(d)-DCIS-immune-microenvironment-spatial-pattern-(Tcell-cluster-formation-follicular-like-vs-diffuse)-(e)-HER2-low-receptor-heterogeneity-spatial-mapping-at-spot-resolution-(f)-paired-companion-validation-of-Xenium-in-situ-single-molecule-findings-at-the-tissue-microenvironment-level-(g)-DCIS-progression-risk-score-validation-at-the-spatial-architecture-level. The 14-paired-DCIS-patient-samples-with-Visium-ST-delivers-complementary-spatial-data-to-the-Xenium-in-situ-atlas-featured-as-id:108-Signal-1-just-4-days-ago-and-extends-the-DCIS-multi-modal-spatial-atlas-coverage-into-the-tissue-microenvironment-spot-based-resolution-needed-for-DCIS-progression-risk-stratification. Therapeutically-actionable-via-DCIS-progression-risk-score-development (Oncotype-DX-breast-DCIS-recurrence-score-validation-at-the-spatial-level) + HER2-low-DCIS-stratification-for-trastuzumab-deruxtecan-T-DXd-Enhertu-DESTINY-Breast06-eligibility + BRCA1-BRCA2-PALB2-germline-carrier-DCIS-stratification-for-olaparib-Lynparza-PARP-inhibitor-adjuvant-therapy. The dataset is ABSENT from the id:101-id:111 retention window per regex-dedup (NO 'GSE348919' / 'Visium-ST' / 'paired-companion-to-id:108' / 'spot-based-spatial' keyword collisions — GSE348584 is in id:108 but GSE348919 is the DISTINCT-Visium-ST-companion-from-the-SAME-paper) and is FIRST promoted to primary feature today as a paired-companion-promotion from id:108.

### 3. Signal 3 (AI Drug Discovery + Protein Variant Effect Prediction Benchmark + ProteinGym Reference Suite + Deep Mutational Scanning DMS Benchmark + OATML-Markslab Canonical Reference + Boltz-2 + ESM-2 + AlphaFold-3 Evaluation) — OATML-Markslab/ProteinGym_v1 (HF dataset, downloads 3122, likes 8, tags size_categories:1M<n<10M + biology + deep-learning + protein-fitness + region:us, score 2, license 'unknown', URL https://huggingface.co/datasets/OATML-Markslab/ProteinGym_v1, canonical-reference-description) + OATML-Markslab/ProteinGym_v0.1 (HF dataset, downloads 2338, likes 15, tags arxiv:2205.13760 + region:us, score 0, URL https://huggingface.co/datasets/OATML-Markslab/ProteinGym_v0.1) + genbio-ai/ProteinGYM-DMS (HF dataset, downloads 1312, likes 1, score 0, URL https://huggingface.co/datasets/genbio-ai/ProteinGYM-DMS, 5-fold cross-validation DMS Benchmark with indels + single substitution + multiple substitutions) + ICML2022/ProteinGym (HF dataset, downloads 363, likes 12, tags arxiv:2205.13760 + region:us, score 0, URL https://huggingface.co/datasets/ICML2022/ProteinGym, original ProteinGym ICML-publication-version with ~1.5M missense variants across 87 DMS assays + ~300k indel mutants across 7 assays) + tyang816/ProteinGym_v1 (HF dataset, downloads 39, likes 0, license apache-2.0, score 2, URL https://huggingface.co/datasets/tyang816/ProteinGym_v1, ProtSSN-Semantical-Topological-Encoding-evaluation-mirror from Tan-2023-bioRxiv-ProtSSN) is the FIRST ProteinGym benchmark cluster to surface in the entire 2026-10-01 hits.jsonl via the 27-query-family fan-out (watcher query 'ProteinGym' [correct routing]) and the FIRST canonical-protein-variant-effect-prediction benchmark to enter the BB digest feed from a HuggingFace-dataset search perspective. ProteinGym (Notin et al. 2023 Nature Methods 'ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design' — https://www.nature.com/articles/s41592-023-02074-9) is THE canonical benchmark suite for protein language model evaluation across 4 sub-benchmarks (zero-shot-substitution-regression / supervised-substitution-regression / zero-shot-indel-regression / supervised-indel-regression) with ~3M total mutations including ~1.5M missense variants across 87 DMS assays + ~300k indel mutants across 7 assays + clinically-annotated-mutations-relevant-to-human-disease + AAV-protein-engineering + Cas9-protein-engineering + therapeutic-protein-optimization + antibody-optimization + biosensor-protein + enzyme-engineering + metagenome-protein + dark-protein evaluation tracks. The 5-HF-artifact-cluster-from-4-distinct-organizations-(OATML-Markslab-=Marks-lab-at-U-of-Toronto-the-canonical-reference-group / genbio-ai-=generative-bio-AI-org-using-5-fold-cross-validation-DMS-benchmark / ICML2022-=original-ProteinGym-ICML-publication-2022-version / tyang816-=ProtSSN-Semantical-Topological-Encoding-evaluation-mirror-by-Tan-Yang-Zhou-Bingxin-2023-bioRxiv) reflects active-community-adoption-and-mirroring-of-the-canonical-benchmark-across-different-research-groups-and-use-cases-(zero-shot-vs-fine-tuned-ProteinGym-evaluation + supervised-ProteinGym-evaluation + protein-design-ProteinGym-evaluation + protein-engineering-ProteinGym-evaluation). The combined-downloads-across-the-cluster-(3122-+-2338-+-1312-+-363-+-39-=-7174-downloads)-is-strong-community-validation-and-establishes-ProteinGym-as-the-gold-standard-benchmark-for-protein-LM-evaluation-in-2026. The-5-HF-artifacts-are-the-reference-suite-for-(a)-Boltz-2-protein-structure-prediction-model-evaluation-(Boltz-2-zero-shot-ProteinGym-regression-benchmark)-(b)-ESM-2-protein-LM-evaluation-(ESM-2-zero-shot-ProteinGym-regression-benchmark)-(c)-AlphaFold-3-protein-structure-prediction-evaluation-(d)-ProteinMPNN-protein-design-evaluation-(inverse-folding-ProteinGym-substitution-benchmark)-(e)-RFdiffusion3-protein-backbone-design-evaluation-(de-novo-protein-design-ProteinGym-evaluation)-(f)-Tranception-ProteinGym-evaluation-(autoregressive-protein-LM-ProteinGym-benchmark)-(g)-ESM-1v-ProteinGym-evaluation-(masked-language-model-protein-ProteinGym-benchmark)-(h)-MSA-Transformer-ProteinGym-evaluation-(i)-ESM-IF-ProteinGym-evaluation-(inverse-folding-protein-LM-ProteinGym-benchmark). Therapeutically-and-scientifically-actionable-as-the-foundation-protein-LM-evaluation-benchmark-for-the-BB-AI-DD-bridge-from-protein-sequence-prediction-to-therapeutic-protein-optimization-including-AAV-capsid-engineering-(gene-therapy-delivery-optimization)-Cas9-protein-engineering-(CRISPR-therapeutic-optimization)-antibody-affinity-maturation-(therapeutic-antibody-optimization)-enzyme-engineering-(industrial-biotech)-metagenome-protein-discovery-(novel-enzyme-discovery)-and-dark-protein-characterization-(function-prediction-for-uncharacterized-proteins). The-dataset-cluster-is-ABSENT-from-the-id:101-id:111-retention-window-per-regex-dedup-(NO-'ProteinGym' / 'OATML-Markslab-ProteinGym' / 'genbio-ai-ProteinGYM-DMS' / 'ICML2022-ProteinGym' / 'tyang816-ProtSSN' / 'protein-variant-effect-prediction-benchmark' / 'deep-mutational-scanning-DMS-benchmark' / 'protein-fitness-prediction' keyword collisions) and is FIRST promoted to primary feature today as a cluster-promotion-from-a-HF-dataset-search.

### 4. Tracker note — Brown Biotech clinical-regulatory + AI-DD + bioinformatics + biotech-infrastructure + multi-omics + longevity: GSE348732 anchors the Fam20C-knockdown / U251-glioma-cell-state / SIBLING-protein-phosphorylation / Golgi-casein-kinase-therapeutic-target / kinase-inhibitor-strategy refractory-GBM lane (BB AI-DD + clinical-translational, with direct Temozolomide-MGMT-resistance + Bevacizumab-anti-VEGF + Optune-TTFields + Regorafenib-Stivarga intersection); GSE348919 anchors the preinvasive-DCIS-Visium-ST-spatial-atlas-companion-to-id:108 / HER2-low-receptor-heterogeneity / DCIS-progression-risk-stratification / Visium-ST-spot-based-spatial-companion-to-Xenium-in-situ-atlas lane (BB bioinformatics + AI-DD + clinical-translational, with direct trastuzumab-deruxtecan-T-DXd-DESTINY-Breast06 + olaparib-Lynparza-BRCA-DCIS + Oncotype-DX-breast-DCIS-recurrence-score intersection); ProteinGym-benchmark-cluster anchors the protein-variant-effect-prediction-benchmark / deep-mutational-scanning-DMS-benchmark / OATML-Markslab-canonical-reference / Boltz-2-ESM-2-AlphaFold-3-ProteinMPNN-RFdiffusion3-evaluation / foundation-model-protein-benchmark AI-DD-biotech-infrastructure lane (BB AI-DD + biotech-infrastructure, with direct Boltz-2-zero-shot-ProteinGym-regression + AAV-capsid-engineering + Cas9-protein-engineering + antibody-affinity-maturation + dark-protein-characterization intersection). The 3 picks span 3 distinct BB lanes (clinical-regulatory-refractory-GBM + bioinformatics-cancer-spatial-DCIS-companion + AI-DD-biotech-infrastructure-protein-LM-benchmark), all ABSENT from the id:101-id:111 retention window per regex-dedup, all freshest-pdat-window-(GSE348732-pdat-2026/09/29-=-1-day-ago-FGSE348919-pdat-2026/09/29-=-1-day-ago-FProteinGym-cluster-=canonical-benchmark-persistent), with a combined 4-axis score (peptide / AI-DD / longevity / cost) of approximately peptide=2/5, AI-DD=5/5, longevity=1/5, cost=5/5 — the AI-DD axis + cost axis both dominate this digest cycle, with the Fam20C-knockdown-U251-glioma-atlas-anchoring-Signal-1-as-the-single-most-actionable-refractory-cancer-therapeutic-target-(score-9-NOVEL-ceiling-break-through-+-1-day-fresh-pdat).

---

## 💼 Next Steps

- **[Open GSE348732 in GEO (Fam20C knockdown U251 glioma transcriptomic atlas, n=6, score 9, pdat 2026/09/29 = 1 day ago)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE348732)**
- **[Open GSE348919 in GEO (preinvasive DCIS Visium ST spatial atlas companion to id:108, n=14, score 6, pdat 2026/09/29 = 1 day ago)](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE348919)**
- **[Open OATML-Markslab/ProteinGym_v1 on HuggingFace (canonical protein-fitness benchmark, downloads 3122, likes 8)](https://huggingface.co/datasets/OATML-Markslab/ProteinGym_v1)**
- **[Open OATML-Markslab/ProteinGym_v0.1 on HuggingFace (original proteinGym v0.1 reference, downloads 2338, likes 15)](https://huggingface.co/datasets/OATML-Markslab/ProteinGym_v0.1)**
- **[Open genbio-ai/ProteinGYM-DMS on HuggingFace (5-fold cross-validation DMS benchmark, downloads 1312, likes 1)](https://huggingface.co/datasets/genbio-ai/ProteinGYM-DMS)**
- **[Open ICML2022/ProteinGym on HuggingFace (original ProteinGym ICML 2022 publication, downloads 363, likes 12)](https://huggingface.co/datasets/ICML2022/ProteinGym)**
- **[Open tyang816/ProteinGym_v1 on HuggingFace (ProtSSN Semantical-Topological-Encoding mirror, downloads 39, apache-2.0)](https://huggingface.co/datasets/tyang816/ProteinGym_v1)**
- **[Read Notin et al. 2023 Nature Methods ProteinGym paper (canonical reference)](https://www.nature.com/articles/s41592-023-02074-9)**
- **[Read Tsuboyama et al. 2023 Nature Mega-scale DMS library paper](https://www.nature.com/articles/s41586-023-06364-6)**
- **[Request Fam20C-knockdown-glioma + SIBLING-protein-phosphorylation + Golgi-casein-kinase-inhibitor brief](https://brownbio.tech/services/ai-drug-discovery#brief)**
- **[Request DCIS-Visium-ST-companion + HER2-low-stratification + DCIS-progression-risk-score brief](https://brownbio.tech/services/biostatx#brief)**
- **[Request ProteinGym-benchmark + Boltz-2-ESM-2-AlphaFold-3-evaluation + protein-LM-benchmark brief](https://brownbio.tech/services/ai-drug-discovery#brief)**
- **[View AI-DD service](https://brownbio.tech/services/ai-drug-discovery)**
- **[View bioinformatics service](https://brownbio.tech/services/biostatx)**
- **[View clinical-translational service](https://brownbio.tech/services/ai-drug-discovery)**
- **[View biotech-infrastructure service](https://brownbio.tech/services/biotech-infrastructure)**
- **[View multiomics service](https://brownbio.tech/services/multiomics)**

---

## 📡 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:** 2026-10-01 06:16 KST
- **Repo:** `ohbryt/brown-biotech-platform`

---

_Auto-published by `brown_biotech_research_digest_publisher.py` · [brownbio.tech](https://brownbio.tech) · Decision-ready research, daily._