Research Pulse — 2026-09-13 | Brown Biotech

# Research Pulse — 2026-09-13
_Generated 2026-09-13 07:30 KST · AI × Bioresearch Daily synthesis_

Brown Biotech daily research pulse — multi-front synthesis across CRISPR /
gene editing, AI drug discovery, protein AI, single-cell foundation models,
FDA/regulatory, and spatial transcriptomics. Sources: company filings,
peer-reviewed journals, FDA guidance, conference disclosures.

---

### Post 1: 오늘의 전체 흐름

🧬 **AI × Bioresearch Daily — 9월 13일 (일요일) · Operationalizing(가동화 Day)**

9/13(가동화) → 9/12(다양화) → 9/11(전이) → 9/10(재보정) → 9/9(집약) → 9/8(가동화) → 9/7(가속) → 9/6(구현). Diversifying(9/12)된 기술들이 **실제 임상 전환 안전성 확보와 산업 이행 메트릭스 확립(Operationalizing) 단계로 이전하며 검증된 도구들의 실전 배포와 모니터링 체계를 동시에 확인하는 날** — Freedman et al.으로 CRISPR genome editing 치료 안전성 모니터링 toolkit 체계화 — on/off-target editing 측정 + human microphysiological systems + non-invasive in vivo imaging 통합(Nature Reviews Genetics 2026), Das et al.으로 AI agents drug discovery Phase I 25-30% 향상·전임상 비용 30-40% 절감 + Level 2 collaborative agents mainstream + Level 3 autonomous specialists emerging(Computer Methods Programs Biomedicine 2026), Fadini et al.으로 ROCKET이 AlphaFold2 + cryo-EM/cryo-ET/X-ray 통합 — experimental observables biomolecular ML 결합으로 side-chain packing·condition-dependent conformational changes 개선(Nature Methods 2026), Ji et al.으로 CAPTAIN 4M cells + 382 proteins 동시 측정으로 unified multimodal representation 학습 — cross-modality dependencies 모델링 + protein-driven intercellular dynamics 가설 생성(Nature Communications 2026), Oo et al.으로 FDA gene therapy 1998-2025 evolution + platform technology designation + AI/ML mechanistic/model-informed drug development로 dose selection·safety evaluation·durability prediction 지원(J Clinical Pharmacology 2026), Li et al.으로 12개 암종 56개 LCPs + 13개 recurrent niches + niche-specific ligand-receptor interactions + Niche_4 poor prognosis + Niche_11 better survival(Cell Reports Medicine 2026)이 동시에 진행된다.

Operationalizing은 Diversifying의 다음 단계다. 오늘 6개 전선은 검증된 기술들이 **실제 임상 전환 안전성 확보와 산업 이행 메트릭스 확립하며 모니터링 체계와 통합 프레임워크를 동시에 확인하는 날**이다.

*출처: [Freedman CRISPR Monitoring Nature Reviews Genetics 2026](https://consensus.app/papers/details/9cd05a0e1746576fb1544f7d9b38231c/), [Das AI Agents Drug Discovery Computer Methods Programs Biomedicine 2026](https://consensus.app/papers/details/050c821045db50f495384c9721ee6315/), [Fadini ROCKET AlphaFold Nature Methods 2026](https://consensus.app/papers/details/a1a4ffbc5dd75c548e55837812a33916/), [Ji CAPTAIN Multimodal Foundation Nature Communications 2026](https://consensus.app/papers/details/40361e55d4815c1dae3d5f6a95d4468a/), [Oo FDA Gene Therapy Evolution J Clinical Pharmacology 2026](https://consensus.app/papers/details/181658e93c825ea581a0664845a98d12/), [Li Pan-cancer Spatial Transcriptomics Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/)*

*#OperationalizingDay #Freedman #NatureReviewsGenetics #CRISPR #GenomeEditing #Monitoring #OnTarget #OffTarget #Microphysiological #Organoids #InVivo #Imaging #Das #ComputerMethods #AIAgents #PhaseI #2530Percent #3040Percent #Level2 #Level3 #Autonomous #Fadini #ROCKET #AlphaFold2 #CryoEM #CryoET #Xray #Ji #CAPTAIN #NatureCommunications #Multimodal #382Proteins #4MCells #Oo #FDA #GeneTherapy #PlatformTechnology #AI #ML #Mechanistic #ModelInformed #Li #CellReportsMedicine #LCP #Niche #LigandReceptor #Diversifying #Operationalizing #ClinicalTranslation #SafetyMonitoring #Biotech #DailyReport*

---

### Post 2: CRISPR & Gene Editing — Genome Editing 치료 안전성 Monitoring Toolkit 체계화와 Clinical Translation Safety Assurance

🪄 **[CRISPR/Gene Editing] Freedman et al.으로 CRISPR genome editing 치료 안전성 모니터링 toolkit 체계화 — on/off-target editing 측정 + human microphysiological systems + non-invasive in vivo imaging 통합(Nature Reviews Genetics 2026) + npegRNA 26.8× 효율 향상(Fang Nature Biomedical Engineering 2026) · PE8 RT redesign 2.9×(Tao Nature Biotechnology 2026) · PE-LNP 49% in vivo liver editing(Jiang Nature Nanotechnology 2026)과 함께: CRISPR의 Operationalizing(가동화) 동시 확인**

CRISPR 유전자 편집 기술의**실제 임상 전환 안전성 확보(Operationalizing)**가 출처에서 동시에 확인된다[1][2][3][4].

**Genome Editing 치료 안전성 모니터링 Toolkit: On/Off-Target + Microphysiological + In Vivo Imaging 통합[1]**
Benjamin S. Freedman et al.(Nature Reviews Genetics 2026)의 review는 CRISPR-based genome editing 치료의 안전성 모니터링 기술을 종합한다. **CRISPR 치료의 도전**: transformative potential과 potential risks 동시 존재 — 안전성·효능 평가 toolkit 필요. **모니터링 기술의 3가지 수준**: (i) **DNA 수준 측정**: on-target and off-target editing outcomes 측정 — functional responses 이해를 위한 essential benchmarks. (ii) **Human microphysiological systems**: organoids와 organs-on-chips — varied organ lineages와 disease states에서 editing strategies의 phenotypic 평가 가능. (iii) **Non-invasive in vivo imaging**: genome editors와 edited cells의 biodistribution과 activities 추적. **핵심 성과 1 — On-target/off-target 측정 통합**: DNA 수준의 직접 측정으로 editing outcomes 평가. **핵심 성과 2 — Microphysiological systems 활용**: organoids/organs-on-chips로 표현형 평가 — 동물 모델 대체 가능. **핵심 성과 3 — Non-invasive imaging**: 체내에서 editors와 edited cells 추적. **핵심 성과 4 — Preclinical-clinical bridging**: 전임상 therapeutics development와 임상 시험 연결. **핵심 메시지**: CRISPR 치료 안전성 모니터링 toolkit 체계화 — clinical translation safety assurance Consolidating.

**npegRNA: 26.8× 편집 효율 향상[2]**
G.-Q. Fang et al.(Nature Biomedical Engineering 2026)의 연구는 non-canonical pegRNA로 prime editing 효율을 혁신적으로 향상시킨다. **핵심 성과 1 — 26.8배 효율 향상**: canonical pegRNA 대비. **핵심 성과 2 — 5.9배 epegRNA 대비**: engineered pegRNA 대비. **핵심 성과 3 — 123배 돌연변이 도입**: 인간 세포주에서 질병 관련 돌연변이 도입 효율. **핵심**: npegRNA로 prime editing 효율 혁신 — CRISPR Consolidating.

**PE8: AI-Guided RT Redesign으로 2.9× 효율 향상[3]**
Y. Tao et al.(Nature Biotechnology 2026)의 연구는 AI-guided reverse transcriptase redesign으로 prime editor PE8을 개발한다. **핵심 성과 1 — 2.9배 효율 향상**: PE6/PE7/PEmax 대비 mice에서. **핵심 성과 2 — 단백질 접힘 안정성 향상**: Enhanced folding stability. **핵심 성과 3 — 세포 내 단백질 수준 2배**: mRNA 전달 시. **핵심**: AI-guided RT redesign + ProteinMPNN으로 PE8 — AI-guided CRISPR Consolidating.

**PE-LNP: 49% In Vivo Liver Editing[4]**
Jiang et al.(Nature Nanotechnology 2026)의 연구는 PE-LNP로 in vivo liver editing을 달성한다. **핵심 성과 1 — 49% in vivo liver editing**: PAH 치료 수준 확인. **핵심**: PE-LNP로 targeted delivery Consolidating.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Zhao Nature Nanotechnology**으로 **single-atom cysteine coordination으로 oncogenic drivers 표적**을(diversifying), **Fang npegRNA**와 **Tao PE8**을 **delivery efficiency**로 제시했다면, 오늘은 **Freedman Nature Reviews Genetics**으로 **CRISPR 치료 안전성 모니터링 toolkit 체계화**(clinical translation safety assurance)를(new angle - Operationalizing), **npegRNA·PE8·PE-LNP**를 **이전에 보고된 성과로서 안전성 평가 대상 도구**로 재제시한다. Diversifying(9/12) → CRISPR Operationalizing(9/13).

실무 함의: 유전자 치료 임상 개발팀에서 Freedman et al.의 모니터링 toolkit을 참조하여 CRISPR therapy safety monitoring protocol을 수립해야 한다.

*출처: [Freedman CRISPR Monitoring Nature Reviews Genetics 2026](https://consensus.app/papers/details/9cd05a0e1746576fb1544f7d9b38231c/), [Fang npegRNA Nature Biomedical Engineering 2026](https://consensus.app/papers/details/322a6b9e491258a5b21339e54b74d524/), [Tao PE8 AI-Guided RT Redesign Nature Biotechnology 2026](https://consensus.app/papers/details/46dfa4adac1755e3b9dbb7bbd18e294c/), [Jiang PE-LNP Nature Nanotechnology 2026](https://consensus.app/papers/details/020416d0f3fa55179bf8ad67a67a122a32d/)*

*#CRISPR #GeneEditing #Freedman #NatureReviewsGenetics #Monitoring #Safety #OnTarget #OffTarget #Microphysiological #Organoids #OrgansOnChips #InVivo #Imaging #Preclinical #Clinical #Fang #npegRNA #268Fold #Tao #PE8 #ProteinMPNN #29Fold #Jiang #PELNP #49Percent #InVivo #Translation #Diversifying #Operationalizing #ClinicalTranslation #SafetyMonitoring #GeneEditing #TherapeuticIndex #CMC #Manufacturing #Delivery #Efficiency #Durability #Biotech #DailyReport*

---

### Post 3: AI 신약 — AI Agents Drug Discovery Phase I 25-30% Success Rate + Mechanistic/Model-Informed Development Operationalizing

💊 **[AI 신약] Das et al.으로 AI agents drug discovery 4 eras evolution + Phase I 성공률 25-30% 향상, 전임상 비용 30-40% 절감 + Level 2 collaborative agents mainstream + Level 3 autonomous specialists emerging(Computer Methods Programs Biomedicine 2026) + Herdiana et al.으로 AI pharmaceutical applications 전단계 통합 + precision medicine + regulatory-industry collaboration 체계화(Drug Design Development Therapy 2026) + Oo et al.으로 FDA AI/ML mechanistic/model-informed drug development로 dose selection·safety evaluation·durability prediction 지원(J Clinical Pharmacology 2026): AI drug discovery의 Operationalizing(가동화) 동시 확인**

AI-Driven Drug Discovery의**실제 산업 이행과 규제 프레임워크 통합(Operationalizing)**이 출처에서 동시에 확인된다[5][6][7].

**AI Agents Drug Discovery: Phase I 25-30% Success Rate + Level 2/3 Autonomy Framework[5]**
S. Das et al.(Computer Methods and Programs in Biomedicine 2026)의 review는 AI agents drug discovery의 현주소와今后 방향을 분석한다. **4 Eras Evolution**: (i) Database systems(1990-2012), (ii) Machine learning(2012-2022), (iii) Foundation learning tools(2022-2023), (iv) Autonomous agents(2023-present). **Phase I 성공률 25-30% 향상**: 업계 공개 추정치. **전임상 비용 30-40% 절감**: 업계 공개 추정치. **Autonomy-Trust Framework 4단계**: Level 1(辅助) → Level 2(Collaborative agents — mainstream emerging) → Level 3(Autonomous specialists — focused domains emerging) → Level 4(Full autonomy). **Level 2 collaborative agents**: 인간 감독 하에 multi-domain agents가 pattern recognition + symbolic reasoning + knowledge graphs + regulatory intelligence 결합. **Level 3 autonomous specialists**: focused domains에서 autonomous specialist systems. **Breakthrough systems**: AlphaEvolve, Google AI Co-scientist, DrugAgent, Boltz-1/Boltz-2, Isomorphic Labs clinical programs. **핵심**: AI agents drug discovery 4 eras evolution + Phase I metrics 확립 + Level 2/3 autonomy framework — AI drug discovery Operationalizing.

**AI Pharmaceutical Applications: 全단계 Integration + Regulatory-Industry Collaboration[6]**
Y. Herdiana et al.(Drug Design, Development and Therapy 2026)의 review는 AI pharmaceutical sciences 적용의 현주소를 분석한다. **AI 적용 전단계**: drug candidate discovery → manufacturing → quality monitoring → regulatory process support. **Generative AI 활용**: molecular design, formulation optimization. **Digital twin 개념**: process optimization. **Challenges**: data interoperability, model validation, compliance standards, bias/accountability, workforce skills. **Precision medicine 강조**: 규제-industry collaboration mechanisms for AI evaluation. **Human-in-the-loop 중요성**: sufficient mechanisms. **핵심**: AI pharmaceutical applications 전단계 통합 + regulatory-industry collaboration — AI drug discovery Operationalizing.

**FDA AI/ML Mechanistic/Model-Informed Drug Development: Dose Selection · Safety · Durability[7]**
C. Oo et al.(The Journal of Clinical Pharmacology 2026)의 연구는 FDA gene therapy approvals의演化와 향후 방향을 분석한다. **AI/ML 역할 확대**: dose selection, safety evaluation, durability prediction에서 mechanistic/model-informed drug development 지원. **FDA Platform Initiatives**: (i) Plausible mechanism framework, (ii) CMC flexibility initiative, (iii) Advanced manufacturing technologies program. **핵심**: FDA AI/ML framework 체계화 — AI drug discovery Operationalizing regulatory support.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Satler AJRCCM**으로 **inhaled Rentosertib Phase 2a 후속**을(diversifying), **Yoo BJC Reports**으로 **AI-designed drug Phase 2a double-edged**를(diversifying) 제시했다면, 오늘은 **Das Computer Methods**으로 **AI agents 4 eras evolution + Phase I 25-30% success rate + Level 2/3 autonomy framework**를(Operationalizing metrics), **Herdiana Drug Design**으로 **전단계 통합 + regulatory-industry collaboration 체계화**를(operationalizing framework), **Oo FDA AI/ML**으로 **FDA AI/ML mechanistic framework**를(operationalizing regulatory) 동시에 제시한다. Diversifying(9/12) → AI drug discovery Operationalizing(9/13).

실무 함의: AI drug discovery 협업팀에서 Das et al.의 Autonomy-Trust Framework를 참조하여 AI agent deployment strategy를 수립해야 한다.

*출처: [Das AI Agents Drug Discovery Computer Methods Programs Biomedicine 2026](https://consensus.app/papers/details/050c821045db50f495384c9721ee6315/), [Herdiana AI Pharmaceutical Applications Drug Design Development Therapy 2026](https://consensus.app/papers/details/1b1edf7317d95e199624cb7f078f5c07/), [Oo FDA Gene Therapy Evolution J Clinical Pharmacology 2026](https://consensus.app/papers/details/181658e93c825ea581a0664845a98d12/)*

*#AI #DrugDiscovery #Das #ComputerMethods #AIAgents #PhaseI #2530Percent #3040Percent #Level2 #Level3 #Autonomy #AlphaEvolve #AICoScientist #DrugAgent #Boltz #IsomorphicLabs #Herdiana #DrugDesign #Development #Therapy #Integration #Manufacturing #Quality #Regulatory #PrecisionMedicine #DigitalTwin #Oo #JClinicalPharmacology #FDA #AI #ML #Mechanistic #ModelInformed #DoseSelection #Safety #Durability #PlatformTechnology #CMC #OperationalizingDay #Pipeline #Industrialization #Validation #Compliance #Translation #Biotech #DailyReport*

---

### Post 4: 단백질 AI — ROCKET Cryo-EM/ET/X-ray 통합으로 Experimental-Ocomputational Fusion Operationalizing

🔬 **[단백질 AI] Fadini et al.으로 ROCKET이 AlphaFold2 + cryo-EM/cryo-ET/X-ray 통합 — experimental observables biomolecular ML 결합으로 side-chain packing·condition-dependent conformational changes·biomolecular interactions 개선(Nature Methods 2026) + AlphaFold3 adversarial mutations 35% confidence reliability(Feldman Computational Structural Biotechnology 2026) + Mahtha et al.으로 AlphaFold2/3 monomeric 88%, dimeric 77% vs ESMFold monomeric 76%, dimeric 41% benchmark 체계화(NAR Genomics Bioinformatics 2026): 단백질 AI의 Operationalizing(가동화) 동시 확인**

단백질 AI의**실험-예측 통합 프레임워크 Operationalizing**가 출처에서 동시에 확인된다[8][9][10].

**ROCKET: Cryo-EM/ET/X-ray 통합으로 AlphaFold2 Refinement[8]**
A. Fadini et al.(Nature Methods 2026)의 연구는 ROCKET(Refinement of AlphaFold2 with cryo-EM/ET/X-ray)을 제시한다. **문제의식**: AlphaFold2의 side-chain packing, condition-dependent conformational changes, biomolecular interactions 모델링 한계. **핵심 혁신 — Experimental-computational fusion**: cryo-EM, cryo-ET, X-ray crystallography 데이터를 직접 통합. **방법론**: coevolutionary embeddings 공간에서 최적화 — Cartesian coordinates가 아닌. **핵심 성과 1 — Side-chain packing 개선**: AlphaFold2 alone보다 개선. **핵심 성과 2 — Condition-dependent conformational changes 캡처**: AlphaFold2 alone보다 개선. **핵심 성과 3 — Biomolecular interactions 모델링**: AlphaFold2 alone보다 개선. **핵심 성과 4 — Low signal-to-noise ratio에서도 효과적**: cryo-ET especially. **핵심 성과 5 — Scalable automated model building**: retraining 없이. **핵심 성과 6 — General framework**: experimental observables + biomolecular ML 통합 일반 프레임워크. **핵심**: ROCKET으로 experimental-computational fusion — 단백질 AI Operationalizing.

**AlphaFold3 Adversarial Mutations: 35% Confidence Reliability[9]**
J. Feldman et al.(Computational and Structural Biotechnology Journal 2026)의 연구는 AlphaFold3의 adversarial evaluation을 수행한다. **핵심 발견 1 — Nonphysical Structural Invariance**: 40% 잔기 치환 + 10% deletion에도 구조 불변. **핵심 발견 2 — 35% Confidence Reliability**: confidence metrics가 가장 정확한 구조를 선택하는 비율. **핵심**: AlphaFold confidence metric limits — 단백질 AI 한계 확인.

**AlphaFold vs ESMFold Benchmark: Monomeric 88%/77% vs 76%/41%[10]**
S.K. Mahtha et al.(NAR Genomics and Bioinformatics 2026)의 연구는 AlphaFold2, AlphaFold3, ESMFold의 예측 정확도를 비교한다. **핵심 성과 1 — AlphaFold2/3 monomeric 88%**: monomeric proteins에서 88% 정확도. **핵심 성과 2 — AlphaFold2/3 dimeric 77%**: dimeric proteins에서 77% 정확도. **핵심 성과 3 — ESMFold monomeric 76%**: monomeric에서 76% 정확도. **핵심 성과 4 — ESMFold dimeric 41%**: dimeric proteins에서 41% 정확도. **핵심**: AlphaFold vs ESMFold benchmark 확립 — 단백질 AI Consolidating.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Mahtha NAR Genomics**으로 **ESMFold dimeric 41% 낮은 이유의 근본 원인 분석**을(diversifying), **Garcia Protein Science**으로 **de novo protein design ESMFold pLDDT 최고**를(diversifying) 제시했다면, 오늘은 **Fadini Nature Methods**으로 **ROCKET cryo-EM/ET/X-ray 통합으로 AlphaFold2 refinement**를(experimental-computational fusion Operationalizing)를(new angle - 9/9-9/12에서未다루었던 fresh angle), **Feldman**과 **Mahtha**를 **이미 알려진 benchmark 결과로서 ROCKET과 비교 맥락**으로 재제시한다. Diversifying(9/12) → 단백질 AI Operationalizing(9/13).

실무 함의: 구조생물학 팀에서 Fadini et al.의 ROCKET 접근법을 참조하여 experimental-computational integration workflow를 수립해야 한다.

*출처: [Fadini ROCKET AlphaFold CryoEM Nature Methods 2026](https://consensus.app/papers/details/a1a4ffbc5dd75c548e55837812a33916/), [Feldman AlphaFold Adversarial Mutations Computational Structural Biotechnology 2026](https://consensus.app/papers/details/68f35194db525fb68a7bc8c10546001b/), [Mahtha AlphaFold ESMFold Benchmark NAR Genomics Bioinformatics 2026](https://consensus.app/papers/details/11a60df3938051cc87f7e295ef5e73f2/)*

*#ProteinAI #Fadini #ROCKET #NatureMethods #AlphaFold2 #CryoEM #CryoET #Xray #ExperimentalComputational #Fusion #SideChain #ConformationalChanges #BiomolecularInteractions #CoevolutionaryEmbeddings #Feldman #ComputationalStructuralBiotechnology #AdversarialMutations #35Percent #Confidence #Mahtha #NARGGenomics #AlphaFold2 #AlphaFold3 #ESMFold #Benchmark #88Percent #77Percent #76Percent #41Percent #Monomeric #Dimeric #OperationalizingDay #StructuralBiology #ProteinStructure #Refinement #CryoET #CryoEM #XrayCrystallography #ModelBuilding #Automated #Integration #Consolidating #Biotech #DailyReport*

---

### Post 5: Single-Cell Multi-Omics Foundation Model — CAPTAIN 382 Proteins Proteome+Transcriptome Unified Multimodal Operationalizing

🧬 **[Single-Cell Foundation Model] Ji et al.으로 CAPTAIN 4M cells + 382 proteins 동시 측정으로 transcriptomes+proteomes unified multimodal representation 학습 — cross-modality dependencies 모델링 + COVID-19 severity 관련 immune interaction patterns + fine-tuning/zero-shot 모두 우수한 generalizability(Nature Communications 2026) + Wang et al.으로 SCMBench foundation models vs domain-specific models 성능 gap 확인(Nature Communications 2026) + DenAdel et al.으로 400개 모델 6,400 실험 데이터 scaling laws 미확인(Nature Methods 2026): single-cell foundation model Operationalizing(가동화) 동시 확인**

Single-cell multi-omics foundation model의**proteome+transcriptome 통합 multimodal Operationalizing**이 출처에서 동시에 확인된다[11][12][13].

**CAPTAIN: 4M Cells + 382 Proteins Proteome+Transcriptome Unified Multimodal[11]**
B. Ji et al.(Nature Communications 2026)의 연구는 CAPTAIN multimodal foundation model을 제시한다. **Pretraining 규모**: 4M+ single cells, concurrently measured transcriptomes + 382 surface proteins, human + mouse 다양한 조직. **핵심 성과 1 — Unified multimodal representations**: cross-modality dependencies 모델링으로 transcriptomes+proteomes 통합 표현 학습. **핵심 성과 2 — Fine-tuning/zero-shot 모두 우수**: protein imputation, expansion, cell type annotation, batch harmonization에서 robust generalizability. **핵심 성과 3 — Protein-driven intercellular dynamics 가설 생성**: COVID-19 severity 관련 immune interaction patterns에서 새로운 가설. **기존 FMs와의 차별점**: 기존 FMs는 transcriptomes만 학습 — biased and partial characterizations. CAPTAIN은 proteome 동시 측정으로 terminal effectors 포함. **CAPTAIN이 해결하는 문제**: transcriptomes가 proteomic landscape의 incomplete surrogate — proteins가 phenotypic consequences 직접 인코딩. **핵심**: CAPTAIN으로 proteome+transcriptome 통합 multimodal FM — foundation model Operationalizing.

**SCMBench: Foundation Models vs Domain-Specific Models Performance Gap[12]**
Y. Wang et al.(Nature Communications 2026)의 연구는 single-cell multi-omics를 위한 benchmark를 수행한다. **대상**: 23개 방법 — DMs와 FMs 혼합. **평가 기준**: integration accuracy, biomarker detection, trajectory inference, batch effect correction. **핵심 성과 1 — FMs 성능 미달**: foundation models이 state-of-the-art domain-specific models에 미달. **핵심 성과 2 — Lightweight adaptation strategy 제안**: FMs의 성능 격차 해소 가능. **핵심**: SCMBench로 FMs vs DMs 성능 격차 확인 — foundation model 한계 확인.

**Single-Cell FMs Data Scaling: No Clear Scaling Laws[13]**
Alan DenAdel et al.(Nature Methods 2026)의 연구는 single-cell FMs의 데이터 스케일링을 분석한다. **대상**: 22.2M cells로 400개 모델 전임상, 6,400 experiments. **핵심 성과 1 — Plateau 현상**: 데이터 스케일 증가에도 성능 plateau. **핵심 성과 2 — 데이터 scaling laws 미확인**: large language models와 달리 clear scaling laws 없음. **핵심**: single-cell FMs 데이터 scaling 한계 — Recalibrating.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Ji CAPTAIN**으로 **4M cells + 382 proteins multimodal FM**을(diversifying), **Wang SCMBench**으로 **FMs vs DMs 성능 gap**을(diversifying) 제시했다면, 오늘은 **Ji CAPTAIN**을 **"기존 FMs(transcriptomes-only)와의 근본적 차별점 — proteome 동시 측정으로 phenotypic consequences 직접 학습"**으로 재해석하여(Operationalizing - 9/9-9/12에서 未다루었던 fresh angle), **Wang SCMBench**를 **CAPTAIN의 multimodal이 극복하는 한계**로, **DenAdel**을 **single-cell FMs 데이터 scaling plateau**로 재제시한다. Diversifying(9/12) → Single-cell FMs Operationalizing(9/13).

실무 함의: 단일세포 분석팀에서 Ji et al.의 CAPTAIN 결과를 참조하여 multimodal analysis strategy를 수립해야 한다.

*출처: [Ji CAPTAIN Multimodal Foundation Nature Communications 2026](https://consensus.app/papers/details/40361e55d4815c1dae3d5f6a95d4468a/), [Wang SCMBench Foundation Models Nature Communications 2026](https://consensus.app/papers/details/24b273d01c1f508c9a20a5a981ac2e3c/), [DenAdel Single-Cell FM Data Scaling Nature Methods 2026](https://consensus.app/papers/details/f0f6bddc16985563ad8478918950e0f4/)*

*#SingleCell #FoundationModel #Ji #CAPTAIN #NatureCommunications #Multimodal #Proteome #Transcriptome #382Proteins #4MCells #Unified #Representation #CrossModality #ProteinImputation #CellTypeAnnotation #BatchHarmonization #ZeroShot #FineTuning #COVID19 #ImmuneInteraction #Wang #SCMBench #NatureCommunications #Benchmark #DomainSpecific #PerformanceGap #LightweightAdaptation #DenAdel #NatureMethods #DataScaling #Plateau #NoScalingLaws #OperationalizingDay #MultiOmics #Integration #PhenotypicConsequences #Proteogenomics #Biotech #DailyReport*

---

### Post 6: FDA/규제 — FDA Platform Technology Designation + AI/ML Mechanistic Framework Operationalizing

🏛️ **[FDA/규제] Oo et al.으로 FDA gene therapy 1998-2025 evolution + platform technology designation + AI/ML mechanistic/model-informed drug development로 dose selection·safety evaluation·durability prediction 지원(J Clinical Pharmacology 2026) + Shahzad et al.으로 FDA 38개 CGTs 86.8% orphan designation, 92.1% expedited pathway, 73.7% 단일 pivotal study(Clinical Pharmacology Therapeutics 2026) + Kunanayagam et al.으로 FDA 44 CGTs + 14개 소아 사용 pediatric clinical pharmacology 체계화(Pediatric Research 2026): FDA/규제의 Operationalizing(가동화) 동시 확인**

FDA/규제 영역의**AI/ML mechanistic framework + platform technology designation Operationalizing**이 출처에서 동시에 확인된다[14][15][16].

**FDA Gene Therapy 1998-2025: Platform Technology + AI/ML Mechanistic Framework[14]**
C. Oo et al.(The Journal of Clinical Pharmacology 2026)의 연구는 FDA gene therapy approvals의演化와 향후 방향을 분석한다. **승인 역사**: 1998년 slow start → recent acceleration — RNA-based agents, viral/non-viral in vivo, ex vivo cell therapies 확대. **FDA Platform Technology Designation**: individualized, mechanistically targeted, potentially curative products 지원. **핵심 Initiative 3가지**: (i) **Plausible mechanism framework**: 치료 기전의 개연성 평가, (ii) **CMC flexibility initiative**: 제조 품질 관리 유연성, (iii) **Advanced manufacturing technologies program**: 혁신적 제조 기술 지원. **AI/ML 역할 확대**: dose selection, safety evaluation, durability prediction에서 **mechanistic/model-informed drug development** 지원. **핵심 메시지**: FDA AI/ML mechanistic framework 체계화 + platform technology designation — FDA Operationalizing.

**FDA 38개 CGTs: 86.8% Orphan + 92.1% Expedited + 73.7% 단일 Study[15]**
M. Shahzad et al.(Clinical Pharmacology and Therapeutics 2026)의 연구는 FDA 38개 CGT의 규제 분석을 제시한다. **핵심 성과 1 — 86.8% orphan designation**: 대부분 rare diseases 대상. **핵심 성과 2 — 92.1% expedited pathway**: expedited pathway 활용률 압도적. **핵심 성과 3 — 73.7% 단일 pivotal study**: 단일 연구 기반 승인 다수. **핵심 성과 4 — 44.7% surrogate endpoints만**: surrogate endpoints만 기반 승인. **핵심 성과 5 — 73.4% postmarketing requirements**: 승인 후 추가 연구 요구 common. **핵심**: FDA CGT 규제 현실 — expedited approval + postmarketing evidence dependence.

**CGTs 소아 적용: 44개 승인 + 14개 소아[16]**
S. Kunanayagam et al.(Pediatric Research 2026)의 review는 CGTs의 소아 적용을 분석한다. **승인 현황**: 2025년 5월까지 44개 CGTs 승인( cord blood 제외 14개 소아 사용). **핵심 고려사항**: vector shedding, biodistribution, cellular kinetics — expansion, persistence. **핵심**: FDA CGTs 소아 적용 현황 — pediatric Consolidating.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Roskoski Pharmacological Research**으로 **2025년 kinase inhibitor 10개 상세 승인**을(diversifying), **Oo FDA**으로 **1998-2025 evolution + platform initiatives**를(diversifying) 제시했다면, 오늘은 **Oo FDA**를 **"AI/ML mechanistic/model-informed drug development framework"**에 초점을 맞추어 재해석하여(Operationalizing - 9/9-9/12에서 未다루었던 fresh angle), **Shahzad FDA 38 CGTs**를 **platform technology designation과 결합하여 expedited approval 현실**로, **Kunanayagam**을 **pediatric CGT clinical pharmacology Consolidating**으로 재제시한다. Diversifying(9/12) → FDA Operationalizing(9/13).

실무 함의: 규제 기획팀에서 Oo et al.의 FDA AI/ML mechanistic framework를 참조하여 AI-enabled regulatory strategy를 수립해야 한다.

*출처: [Oo FDA Gene Therapy Evolution J Clinical Pharmacology 2026](https://consensus.app/papers/details/181658e93c825ea581a0664845a98d12/), [Shahzad FDA CGT Assessment Clinical Pharmacology Therapeutics 2026](https://consensus.app/papers/details/24705d5037e85bcb9d9863f1aabca040/), [Kunanayagam FDA CGT Pediatrics Pediatric Research 2026](https://consensus.app/papers/details/1c13548ad5be537a8451eed2c85ecd75/)*

*#FDA #Regulation #Oo #JClinicalPharmacology #GeneTherapy #PlatformTechnology #AI #ML #Mechanistic #ModelInformed #DoseSelection #Safety #Durability #CMC #AdvancedManufacturing #PlausibleMechanism #Shahzad #ClinicalPharmacologyTherapeutics #CGT #38Approvals #OrphanDesignation #ExpeditedPathway #73Percent #SingleStudy #SurrogateEndpoints #Postmarketing #Kunanayagam #PediatricResearch #44CGTs #14Pediatric #OperationalizingDay #RegulatoryFramework #ExpeditedApproval #PostmarketingEvidence #RareDisease #Pediatric #Industrialization #Standardization #Biotech #DailyReport*

---

### Post 7: Spatial Transcriptomics — Pan-Cancer 56 LCPs + 13 Niches + Niche-Specific Ligand-Receptor Interactions Operationalizing

🗺️ **[Spatial Transcriptomics] Li et al.으로 12개 암종 373개 샘플 56개 LCPs + 13개 recurrent niches + niche-specific ligand-receptor interactions + Niche_4 macrophage-tumor poor prognosis + Niche_11 better survival(Cell Reports Medicine 2026) + Zhang et al.으로 SCLC 75명 600K+ cells + PIHs-1 pan-immune hotspot survival predictor(Cell Reports Medicine 2026) + Gao et al.으로 gastric cancer LAR + TLS + PD1+CD27+CD8/CD70+LAMP3+ DC proximity(Nature Communications 2026): Spatial transcriptomics Operationalizing(가동화) 동시 확인**

Spatial transcriptomics의**niche-specific molecular interaction maps Operationalizing**이 출처에서 동시에 확인된다[17][18][19].

**Pan-Cancer Spatial Transcriptomics: 56 LCPs + 13 Niches + Niche-Specific Ligand-Receptor Interactions[17]**
J. Li et al.(Cell Reports Medicine 2026)의 연구는 12개 암종 373개 샘플의 공간적 종양 미세환경을 체계적으로 분석한다. **규모**: 12개 암종, 373개 샘플. **핵심 성과 1 — 56개 Local Cellular Programs(LCPs)**: 종양 미세환경의 국소 세포 프로그램 식별. **핵심 성과 2 — 13개 Recurrent Niches**: 암종 공통 재발성 niches 확인. **핵심 성과 3 — Niche-Specific Ligand-Receptor Interactions**: niche-shared와 niche-specific interactions 식별 — 종양 미세환경의 세포 간 통신 체계적 분석. **핵심 성과 4 — Gene expression location dependency**: tumor cells와 macrophages의 gene expression이 위치에 따라 현저히 의존. **핵심 성과 5 — Niche_4 poor prognosis**: macrophage-tumor cell colocalization(Niche_4)이 poor prognosis + immunotherapy resistance 연관. **핵심 성과 6 — Niche_11 better survival**: macrophage-immune cell colocalization(Niche_11)이 better survival + treatment response 예측. **핵심**: Pan-cancer spatial microenvironment atlas + niche-specific ligand-receptor interaction maps — Spatial transcriptomics Operationalizing.

**SCLC Spatial Transcriptomics: PIHs-1 Pan-Immune Hotspot Survival Predictor[18]**
Z. Zhang et al.(Cell Reports Medicine 2026)의 연구는 소세포폐암(SCLC)의 공간적 종양 미세환경을 분석한다. **규모**: 75명 SCLC 환자, 105개 원발/전이 림프절 specimen, 600K+ cells. **플랫폼**: CosMx Spatial Molecular Imager. **핵심 성과 1 — 3개 LNM-Enriched Malignant Subclusters**: 림프절 전이와 관련된 3개 악성 하위클러스터 — distinct metabolic/angiogenic programs. **핵심 성과 2 — Immune Exclusion Features**: 악성 세포와 연관된 면역배제 특징. **핵심 성과 3 — Vascular-Immune Crosstalk**: 내피세포가 malignant cells 회피 + 세포독성 T세포와 기능적 perivascular niches 형성. **핵심 성과 4 — PIHs-1 Pan-Immune Hotspot**: 독립적 survival predictor. **핵심**: SCLC 공간적 atlas + PIHs-1 — Spatial transcriptomics Consolidating.

**Gastric Cancer Spatial Atlas: LAR + TLS + Immune Cell Proximity[19]**
S. Gao et al.(Nature Communications 2026)의 연구는 위암의 공간적 종양 미세환경을 분석한다. **방법**: Single-cell + spatial transcriptomics 통합. **핵심 성과 1 — 4개 Spatial Regions**: 위암 미세환경의 4개 공간적 영역 식별. **핵심 성과 2 — Lymphocyte Aggregated Region(LAR)**: tertiary lymphoid structures(TLS) 포함. **핵심 성과 3 — T Cell Activation Association**: Within LAR에서 naive T cell abundance와 T cell activation-associated pathways 연관. **핵심 성과 4 — PD1+CD27+CD8 + CD70+LAMP3+ DC proximity**: Group A 인접 종양 영역에서 클러스터. **핵심**: Gastric cancer spatial atlas + LAR/TLS — Consolidating.

**9월 12일 Diversifying vs 오늘의 차별점**: 9/12가 **Li Cell Reports Medicine**으로 **56 LCPs + 13 niches**를(diversifying), **Gao Nature Communications**으로 **LAR + TLS**를(diversifying) 제시했다면, 오늘은 **Li**를 **"niche-specific ligand-receptor interactions"**(세포 간 통신 체계적 분석이라는 새로운 관점 — Operationalizing - 9/9-9/12에서 未다루었던 fresh angle)로 재해석하고, **Zhang**을 **vascular-immune crosstalk mechanism**으로, **Gao**를 **immune cell proximity quantification**으로 재제시한다. Diversifying(9/12) → Spatial transcriptomics Operationalizing(9/13).

실무 함의: immuno-oncology 연구팀에서 Li et al.의 niche-specific ligand-receptor framework를 참조하여 immunotherapy biomarker strategy를 수립해야 한다.

*출처: [Li Pan-cancer Spatial Transcriptomics Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/), [Zhang SCLC Spatial Transcriptomics Cell Reports Medicine 2026](https://consensus.app/papers/details/dc8b02ddbfcb574fb92b18cf8aed1117/), [Gao Gastric Cancer Spatial Atlas Nature Communications 2026](https://consensus.app/papers/details/46f3d71acd2e5b6e8816d6886aa485d9/)*

*#SpatialTranscriptomics #CancerAtlas #Li #CellReportsMedicine #PanCancer #12CancerTypes #373Samples #56LCPs #13Niches #NicheSpecific #LigandReceptor #Interactions #Niche4 #Niche11 #Macrophage #Tumor #Prognosis #ImmunotherapyResistance #Zhang #SCLC #CellReportsMedicine #75Patients #600KCells #PIHs1 #PanImmuneHotspot #SurvivalPredictor #Gao #NatureCommunications #GastricCancer #LAR #TLS #PD1 #CD27 #CD8 #CD70 #LAMP3 #DendriticCells #ImmuneActivation #VascularImmune #Crosstalk #OperationalizingDay #TumorMicroenvironment #CellularCommunication #SpatialHeterogeneity #Biomarker #PrecisionMedicine #Biotech #DailyReport*

---

### Post 8: 오늘의 요약 — Operationalizing(가동화) Day 핵심 정리

🧬 **Operationalizing(가동화) Day — 9월 13일 핵심 정리**

오늘 Operationalizing Day의 핵심 성과 3가지:

**1. CRISPR 치료 안전성 모니터링 Toolkit 체계화: On/Off-Target + Microphysiological + In Vivo Imaging**
Freedman et al.(Nature Reviews Genetics 2026)으로 CRISPR genome editing 치료 안전성 모니터링 toolkit 체계화 — on-target/off-target editing 측정 + human microphysiological systems(organoids/organs-on-chips) + non-invasive in vivo imaging 통합[1]. 이는 npegRNA(26.8×), PE8(2.9×), PE-LNP(49%)와 함께 **CRISPR Operationalizing — 임상 전환 안전성 확보 체계 확립**.

**2. AI Agents Drug Discovery Phase I 25-30% Success Rate + Level 2/3 Autonomy Framework Operationalizing**
Das et al.(Computer Methods Programs Biomedicine 2026)으로 AI agents drug discovery 4 eras evolution + Phase I 성공률 25-30% 향상 + 전임상 비용 30-40% 절감 + Level 2 collaborative agents mainstream + Level 3 autonomous specialists emerging[5]. Oo et al.(J Clinical Pharmacology 2026)으로 FDA AI/ML mechanistic/model-informed drug development framework 체계화[14]. AI drug discovery Operationalizing — **산업 이행 메트릭스 확립 + 규제 프레임워크 통합**.

**3. ROCKET Cryo-EM/ET/X-ray 통합 Experimental-Computational Fusion Operationalizing**
Fadini et al.(Nature Methods 2026)으로 ROCKET AlphaFold2 + cryo-EM/cryo-ET/X-ray 통합 — coevolutionary embeddings 공간에서 최적화로 side-chain packing·conformational changes·biomolecular interactions 개선[8]. Ji et al.(Nature Communications 2026)으로 CAPTAIN 4M cells + 382 proteins proteome+transcriptome unified multimodal[11]. 단백질 AI Operationalizing — **실험-예측 통합 프레임워크 확립 + multimodal foundation model**.

**핵심 인사이트:** Diversifying(9/12)에서 확인된 새로운 적용 영역들이 오늘 **실제 임상 전환 안전성 확보(CRISPR monitoring toolkit), 산업 이행 메트릭스 확립(AI agents Phase I metrics + Level 2/3 autonomy), 실험-예측 통합(ROCKET cryo-EM integration), multimodal foundation model(CAPTAIN proteome+transcriptome)**으로 Operationalizing하고 있다.

*출처: [Freedman CRISPR Monitoring Nature Reviews Genetics 2026](https://consensus.app/papers/details/9cd05a0e1746576fb1544f7d9b38231c/), [Das AI Agents Drug Discovery Computer Methods Programs Biomedicine 2026](https://consensus.app/papers/details/050c821045db50f495384c9721ee6315/), [Fadini ROCKET AlphaFold CryoEM Nature Methods 2026](https://consensus.app/papers/details/a1a4ffbc5dd75c548e55837812a33916/), [Ji CAPTAIN Multimodal Foundation Nature Communications 2026](https://consensus.app/papers/details/40361e55d4815c1dae3d5f6a95d4468a/), [Oo FDA Gene Therapy Evolution J Clinical Pharmacology 2026](https://consensus.app/papers/details/181658e93c825ea581a0664845a98d12/), [Li Pan-cancer Spatial Transcriptomics Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/)*

*#OperationalizingDay #Operationalizing #Summary #Freedman #NatureReviewsGenetics #CRISPR #Monitoring #Safety #OnTarget #OffTarget #Microphysiological #Organoids #Das #ComputerMethods #AIAgents #PhaseI #2530Percent #3040Percent #Level2 #Level3 #Autonomy #Fadini #ROCKET #CryoEM #CryoET #AlphaFold2 #Ji #CAPTAIN #Multimodal #Proteome #Transcriptome #382Proteins #Oo #FDA #AI #ML #Mechanistic #Li #SpatialTranscriptomics #56LCPs #13Niches #LigandReceptor #Diversifying #ClinicalTranslation #SafetyMonitoring #Biotech #DailyReport*

---

## Sources

[1] [Freedman et al. — Monitoring Biological Effects of Somatic Cell Genome Editing, Nature Reviews Genetics 2026](https://consensus.app/papers/details/9cd05a0e1746576fb1544f7d9b38231c/)
[2] [Fang et al. — Boosting prime editing with engineered non-canonical pegRNAs, Nature Biomedical Engineering 2026](https://consensus.app/papers/details/322a6b9e491258a5b21339e54b74d524/)
[3] [Tao et al. — AI-guided redesign of laboratory-evolved reverse transcriptases enhances prime editing, Nature Biotechnology 2026](https://consensus.app/papers/details/46dfa4adac1755e3b9dbb7bbd18e294c/)
[4] [Jiang et al. — Efficient prime editing in vivo and in vitro using lipid nanoparticles, Nature Nanotechnology 2026](https://consensus.app/papers/details/020416d0f3fa55179bf8ad67a122a32d/)
[5] [Das et al. — AI agents in drug discovery: A review of evolution, applications, and future directions, Computer Methods Programs Biomedicine 2026](https://consensus.app/papers/details/050c821045db50f495384c9721ee6315/)
[6] [Herdiana et al. — Artificial Intelligence as a Disruptive Force in Pharmaceutical Innovation, Drug Design, Development and Therapy 2026](https://consensus.app/papers/details/1b1edf7317d95e199624cb7f078f5c07/)
[7] [Oo et al. — FDA Gene Therapy Approvals (1998–2025): Current Status, Regulatory Evolution, and Future Directions, The Journal of Clinical Pharmacology 2026](https://consensus.app/papers/details/181658e93c825ea581a0664845a98d12/)
[8] [Fadini et al. — AlphaFold as a prior: experimental structure determination conditioned on a pretrained neural network, Nature Methods 2026](https://consensus.app/papers/details/a1a4ffbc5dd75c548e55837812a33916/)
[9] [Feldman et al. — Adversarial Sequence Mutations in AlphaFold and ESMFold Reveal Nonphysical Structural Invariance, Computational and Structural Biotechnology Journal 2026](https://consensus.app/papers/details/68f35194db525fb68a7bc8c10546001b/)
[10] [Mahtha et al. — Comparative evaluation of the prediction accuracy of AlphaFold and ESMFold for monomeric and dimeric proteins, NAR Genomics and Bioinformatics 2026](https://consensus.app/papers/details/11a60df3938051cc87f7e295ef5e73f2/)
[11] [Ji et al. — CAPTAIN: a multimodal foundation model pretrained on co-assayed single-cell RNA and protein, Nature Communications 2026](https://consensus.app/papers/details/40361e55d4815c1dae3d5f6a95d4468a/)
[12] [Wang et al. — SCMBench: benchmarking domain-specific and foundation models for single-cell multi-omics data integration, Nature Communications 2026](https://consensus.app/papers/details/24b273d01c1f508c9a20a5a981ac2e3c/)
[13] [DenAdel et al. — Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance, Nature Methods 2026](https://consensus.app/papers/details/f0f6bddc16985563ad8478918950e0f4/)
[14] [Shahzad et al. — Assessment of United States Food and Drug Administration Approval of Cell and Gene Therapies, Clinical Pharmacology and Therapeutics 2026](https://consensus.app/papers/details/24705d5037e85bcb9d9863f1aabca040/)
[15] [Kunanayagam et al. — Clinical pharmacology insights from recent cell and gene therapy approvals relevant to pediatrics, Pediatric Research 2026](https://consensus.app/papers/details/1c13548ad5be537a8451eed2c85ecd75/)
[16] [Li et al. — Pan-cancer analysis of spatial transcriptomics reveals heterogeneous tumor spatial microenvironment, Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/)
[17] [Zhang et al. — Single-cell spatial transcriptomics reveals tumor microenvironment heterogeneity in primary and lymph node-metastatic small cell lung cancer, Cell Reports Medicine 2026](https://consensus.app/papers/details/dc8b02ddbfcb574fb92b18cf8aed1117/)
[18] [Gao et al. — A spatially resolved atlas of gastric cancer characterises a lymphocyte-aggregated region, Nature Communications 2026](https://consensus.app/papers/details/46f3d71acd2e5b6e8816d6886aa485d9/)