Research Pulse — 2026-10-21 | Brown Biotech

# Research Pulse — 2026-10-21
_Generated 2026-10-21 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 — 10월 21일 (수요일) · Deepening(심화) Day**

10/21(심화) → 10/17(임상실현) → 10/16(연결) → 10/15(규모화) → 10/14(평가) → 10/13(표준화) → 10/12(종합) → 10/11(문맥화). Clinicalizing(10/17)된 Framework들이 **각 분야主流 체계에서 깊이 있는 메커니즘 이해·검증 체계 확립·한계 인식 심화(Deepening)되는 날** — Jong et al.로 Prime Editing PE7 review: PE1→PE7 evolution + editing efficiency + in vivo delivery + "CRISPR 3.0" platform(BMB Reports 2026), Chakraborty et al.로 CRISPR therapeutics approval landscape: Casgevy UK+MHRA+FDA + exa-cel + lovo-cel + industry-academia collaboration + patent landscape(Molecular Therapy Nucleic Acids 2026), Magdy et al.로 CRISPR on/off-target activities: exagamglogene autotemcel + kayjayguran abengcemeran + PAM-altered base editing + on-target/off-target spectrum(Trends Biotechnology 2026), Ock et al.로 LLM Agent modular framework: molecular generation + 75 properties + QED >0.6 34→55 + Boltz-2 + binding affinity + scalable drug discovery(Journal Chemical Information Modeling 2025), Huynh et al.로 AI agents drug discovery: autonomous reasoning + literature synthesis + toxicity prediction + molecular design + end-to-end decision-making(Drug Discovery Today 2026), Malhotra et al.로 AlphaFold comprehensive review: AF1→AF2→AF3 evolution + drug discovery applications + vaccine design + dynamics limitations(Computers Biology Medicine 2025), Kedzierska et al.로 scGPT/Geneformer zero-shot limitations: 148 citations + reliability challenges + simpler methods outperform + discovery settings(Kedzierska Genome Biology 2025), Li et al.로 pan-cancer spatial microenvironment: 373 samples + 12 cancer types + 56 LCPs + 13 niches + Niche_4 poor prognosis + Niche_11 better survival(Cell Reports Medicine 2026)이 동시에 진행된다.

Deepening은 Clinicalizing의 다음 단계다. 오늘 6개 전선은 임상 결과가 나온 Framework들이 **실제 메커니즘 심화·검증 체계 확립·한계 인식 심화(Deepening)되는 날**이다.

*출처: [Jong Prime Editing PE7 BMB Reports 2026](https://consensus.app/papers/details/11216802c06153dca45549e3d4ff98f5/), [Chakraborty CRISPR Therapeutics Landscape Molecular Therapy Nucleic Acids 2026](https://consensus.app/papers/details/68148f0062ec554ba4af1996dd3718cb/), [Magdy CRISPR On/Off-Target Trends Biotechnology 2026](https://consensus.app/papers/details/2952c798ad0458e188c7cc56cffc4dec/), [Ock LLM Agent Modular Journal Chemical Information Modeling 2025](https://consensus.app/papers/details/2332c070569c5b5385f063ecacff1003/), [Huynh AI Agents Drug Discovery Today 2026](https://consensus.app/papers/details/d47b0cf47a4f5055a7175a6fe09bb6e4/), [Malhotra AlphaFold Comprehensive Review Computers Biology Medicine 2025](https://consensus.app/papers/details/97539d9282a55cc7b207d41360287063/), [Kedzierska scGPT Geneformer Zero-Shot Genome Biology 2025](https://consensus.app/papers/details/1574cad890b251c7af78919561e7d8d1/), [Li Pan-Cancer Spatial Microenvironment Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/)*

*#DeepeningDay #Jong #BMBReports #PrimeEditing #PE7 #PE1ToPE7 #CRISPR3Point0 #InvivoDelivery #Chakraborty #MolecularTherapy #NucleicAcids #Casgevy #ExaCel #LovoCel #ExpeditedApproval #Magdy #TrendsBiotechnology #OnTarget #OffTarget #PAMAltered #BaseEditing #Ock #JournalChemicalInformation #LLMAgent #ModularFramework #MolecularGeneration #Boltz2 #BindingAffinity #Huynh #DrugDiscoveryToday #AIAgents #AutonomousReasoning #LiteratureSynthesis #ToxicityPrediction #Malhotra #ComputersBiologyMedicine #AlphaFold #AF2 #AF3 #Evolution #VaccineDesign #Kedzierska #GenomeBiology #scGPT #Geneformer #ZeroShot #Reliability #Li #CellReportsMedicine #SpatialMicroenvironment #12CancerTypes #56LCPs #13Niches #Niche4 #Niche11 #Deepening #MechanisticUnderstanding #Validation #BiomedicalAI #DailyReport*

---

### Post 2: CRISPR & Gene Editing — Prime Editing PE7 Evolution × On/Off-Target Spectrum Deepening

🪄 **[CRISPR/Gene Editing] Jong et al.로 Prime Editing PE7 review: PE1→PE7 evolution + editing efficiency + in vivo delivery + "CRISPR 3.0" platform(BMB Reports 2026) + Magdy et al.로 CRISPR on/off-target activities: exagamglogene autotemcel + kayjayguran abengcemeran + PAM-altered base editing + on-target/off-target spectrum optimization(Trends Biotechnology 2026) + Chakraborty et al.로 CRISPR therapeutics approval landscape: Casgevy UK/MHRA/FDA + exa-cel + lovo-cel + industry-academia collaboration + patent landscape(Molecular Therapy Nucleic Acids 2026): CRISPR Deepening(심화) 동시 확인**

CRISPR gene therapy의 **PE7 evolution + on/off-target spectrum 심화(Deepening)**이 출처에서 동시에 확인된다[1][2][3].

**Prime Editing PE7: PE1→PE7 Evolution + "CRISPR 3.0" Platform[1]**
Uijin Jong et al.(BMB Reports 2026)의 review는 prime editing의 PE1에서 PE7까지의 방법론적 진화를 종합한다[1]. **PE1 to PE7**: 2019년 이후 지속적 효율성 개선. **Editing efficiency improvement**: 역반적효소(RT) 엔지니어링 + Cas9 nickase 최적화. **In vivo delivery technologies**: 나노입자 + 분할 바이러스 시스템. **"CRISPR 3.0" platform**: 정밀 유전자 치료의 차세대 플랫폼으로 자리매김. **Large genomic region correction**: 대규모 유전체 교정 능력 확대. **Editable target range expansion**: 편집 가능 표적 범위 확장. **PE7 advantages over conventional CRISPR-Cas9**: double-strand break 없이 모든 유형의 점변이 가능. 핵심: Prime editing PE7 — 이행의 방법론 심화(Deepening).

**CRISPR On/Off-Target Spectrum: exagamglogene + kayjayguran + PAM-Altered Base Editing[2]**
Magdy et al.(Trends in Biotechnology 2026)의 opinion article은 FDA 승인 CRISPR 치료제의 분자적 스펙트럼을 분석한다[2]. **exagamglogene autotemcel**: ex vivo hemoglobinopathies 치료용 CRISPR nuclease Cas9. **kayjayguran abengcemeran**: in vivo 초희귀 대사 질환용 PAM-altered base-editing Cas9 variant. **PAM-altered base editing**: 전통적 CRISPR의 PAM 제한 극복. **On-target vs off-target spectrum**: 치료 효과와 부작용의 분자적 원인 이해. **CRISPR biotechnology spectrum**: 단순 in vitro 도구에서 치료제로 이행. **"On-target" molecular variations**: 표적 유전자 변이의 치료적 의미. **"Off-target" unintended modifications**: 안전성 프로파일的核心. 핵심: CRISPR on/off-target — 이행의 분자적 심화(Deepening).

**CRISPR Therapeutics Approval Landscape: Casgevy + exa-cel + lovo-cel + Patent Landscape[3]**
Chakraborty et al.(Molecular Therapy. Nucleic Acids 2026)의 review는 CRISPR 치료제의 승인 현황과 특허地貌를 종합한다[3]. **Casgevy UKMHRA November 16, 2023**: SCD 치료제 최초 승인. **Casgevy FDA December 8, 2023**: 미국 FDA 승인. **exa-cel**: sickle cell disease 치료용. **lovo-cel**: 추가适应증 확장. **Industry-academia collaboration**: 제약-학술 협업 모델. **Patent landscape**: CRISPR-Cas9 특허 체계. **Challenges + safety concerns + cost-effectiveness**: 승인 후 과제. **Countries' policymakers + biopharmaceutical companies + next-generation researchers**: 이해관계자 폭넓음. 핵심: CRISPR therapeutics approval landscape — 이행의 규제 현실 심화(Deepening).

**10월 17일 Clinicalizing(Cohn NEJM Lonvoguran -87% + Shaik 148 patients + Sheikh Kresladi) vs 오늘의 차별점**: 10/17이 **Cohn NEJM**으로 Lonvoguran Phase 3 RCT -87% attack rate를, **Shaik European Journal**으로 SCD gene therapy 148 patients 9 programs delivery outcomes를, **Sheikh Annals**으로 Kresladi FDA March 2026 approval을 제시했다면, 오늘은 **Jong BMB Reports**으로 **PE1→PE7 evolution + in vivo delivery + "CRISPR 3.0"**이라는 방법론 심화(Deepening)를, **Magdy Trends Biotechnology**으로 **exagamglogene + kayjayguran + PAM-altered base editing의 on/off-target molecular spectrum**이라는 분자적 심화(Deepening)를, **Chakraborty Molecular Therapy**으로 **Casgevy UK/MHRA/FDA + exa-cel + lovo-cel + patent landscape의 규제-산업 현실 심화**(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → PE7 methodology + molecular spectrum + approval landscape Deepening(10/21).

실무 함의: 유전자 치료 개발팀에서 Jong et al.의 PE7 evolution을 next-generation prime editing strategy에 참조해야 한다.

*출처: [Jong Prime Editing PE7 BMB Reports 2026](https://consensus.app/papers/details/11216802c06153dca45549e3d4ff98f5/), [Magdy CRISPR On/Off-Target Activities Trends Biotechnology 2026](https://consensus.app/papers/details/2952c798ad0458e188c7cc56cffc4dec/), [Chakraborty CRISPR Therapeutics Approval Landscape Molecular Therapy Nucleic Acids 2026](https://consensus.app/papers/details/68148f0062ec554ba4af1996dd3718cb/)*

*#CRISPR #GeneEditing #Jong #BMBReports #PrimeEditing #PE7 #PE1ToPE7 #CRISPR3Point0 #InvivoDelivery #Efficiency #Magdy #TrendsBiotechnology #OnTarget #OffTarget #Exagamglogene #Kayjayguran #PAMAltered #BaseEditing #MolecularSpectrum #Chakraborty #MolecularTherapy #NucleicAcids #Casgevy #UKMHRA #FDA #ExaCel #LovoCel #IndustryAcademia #PatentLandscape #SafetyConcerns #CostEffectiveness #DeepeningDay #MethodologicalDeepening #MolecularUnderstanding #ApprovalReality #BiomedicalAI #DailyReport*

---

### Post 3: AI 신약 — LLM Agent Modular Framework × AI Agents Clinical Translation: Deepening

💊 **[AI 신약] Ock et al.로 LLM Agent modular framework: molecular generation + 75 properties + QED >0.6 34→55 + Boltz-2 + binding affinity + scalable drug discovery(Journal Chemical Information Modeling 2025) + Huynh et al.로 AI agents drug discovery: autonomous reasoning + literature synthesis + toxicity prediction + molecular design + end-to-end decision-making(Drug Discovery Today 2026) + Zheng et al.로 LLM drug discovery review: paradigm shift + disease mechanisms + de novo drug discovery + clinical trial optimization(Patterns 2025): AI Drug Discovery Deepening(심화) 동시 확인**

AI-Driven Drug Discovery의 **LLM agent modular framework + AI agents clinical translation 심화(Deepening)**가 출처에서 동시에 확인된다[4][5][6].

**LLM Agent Modular Framework: Molecular Generation + 75 Properties + QED 34→55 + Boltz-2[4]**
Janghoon Ock et al.(Journal of Chemical Information and Modeling 2025)의 연구는 LLM 기반 модуляр 프레임워크를 제시한다[4]. **Modular framework powered by LLMs**: domain-specific tools 통합. **Biomedical data retrieval**: 생물의학적 데이터 검색. **Literature-grounded Q&A via RAG**: 검색 증강 생성. **Molecular generation**: 분자 생성. **Multi-property prediction**: 75 properties 예측. **ADMET + physicochemical descriptors**: 흡수/분포/대사/배설/독성 특성. **Property-aware molecular refinement**: 특성 인식 분자 정제. **QED >0.6: 34→55 molecules**: 2 라운드 정제 후 100개 중 34개→55개. **Boltz-2**: 3D protein-ligand 구조 생성. **Binding affinity estimation**: 결합 친화도 예측. **Scalable foundation**: 확장 가능한 기반. 핵심: LLM agent modular — 이행의 기술 체계 심화(Deepening).

**AI Agents Drug Discovery: Autonomous Reasoning + Literature + Toxicity + End-to-End[5]**
Dinh-Long Huynh et al.(Drug Discovery Today 2026)의 review는 AI agents의 drug discovery 적용을 분석한다[5]. **LLM-based autonomous reasoning**: large language model 기반 자율 추론. **Literature synthesis**: 문헌 종합. **Toxicity prediction**: 독성 예측. **Molecular design**: 분자 설계. **Automated protocol generation**: 자동화된 프로토콜 생성. **End-to-end decision-making**: 종단간 의사결정. **Early implementations**: substantial gains in speed, reproducibility, scalability. **Challenges**: data heterogeneity + system reliability + privacy + benchmarking. 핵심: AI agents drug discovery — 이행의 개념 심화(Deepening).

**LLM Drug Discovery Review: Paradigm Shift + Disease Mechanisms + Clinical Trials[6]**
Yi Zheng et al.(Patterns 2025)의 review는 LLM의 drug discovery paradigm shift를 종합한다[6]. **Paradigm shift in drug development**: 계산 방법론의 근본적 전환. **Disease mechanisms understanding**: 질병 메커니즘 이해 증진. **De novo drug discovery**: 새로운 약물 발견. **Clinical trial optimization**: 임상 시험 최적화. **Target-disease linkage**: 표적-질병 연관성 해석. **Complex biomedical data interpretation**: 복잡한 생물의학적 데이터 처리. **Drug efficacy + safety prediction**: 약물 효능/안전성 예측. **Computational biology + pharmacology + AI4Science**: 융합 영역. 핵심: LLM drug discovery — 이행의 패러다임 심화(Deepening).

**10월 17일 Clinicalizing(Mao Phase IIa + Seyhan clinical candidates + Cai preclinical pipeline) vs 오늘의 차별점**: 10/17이 **Mao Frontiers Pharmacology**으로 Phase IIa encouraging + Phase I safety signals를, **Seyhan 재확인**으로 scaffold-aware tools + 0건 완전 마케팅 승인을, **Cai Pharmaceuticals**으로 multimodal + digital twins를 제시했다면, 오늘은 **Ock Journal Chemical Information**으로 **LLM agent modular framework + 75 properties + QED >0.6 34→55 + Boltz-2 binding affinity**라는 구체적 수치 기반 기술 심화(Deepening)를, **Huynh Drug Discovery Today**으로 **autonomous reasoning + literature synthesis + toxicity + end-to-end decision-making**의 AI agent 개념 심화(Deepening)를, **Zheng Patterns**으로 **disease mechanisms + de novo drug discovery + clinical trial optimization의 paradigm shift** 심화(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → modular framework 수치 + AI agent 개념 + paradigm shift 심화 Deepening(10/21).

실무 함의: AI drug discovery 협업팀에서 Ock et al.의 modular framework를 AI-assisted molecular design pipeline에 참조해야 한다.

*출처: [Ock LLM Agent Modular Framework Journal Chemical Information Modeling 2025](https://consensus.app/papers/details/2332c070569c5b5385f063ecacff1003/), [Huynh AI Agents Drug Discovery Drug Discovery Today 2026](https://consensus.app/papers/details/d47b0cf47a4f5055a7175a6fe09bb6e4/), [Zheng LLM Drug Discovery Patterns 2025](https://consensus.app/papers/details/5abbd25ca9d65110a7dacde7c5455b2e/)*

*#AI #DrugDiscovery #Ock #JournalChemicalInformation #ModularFramework #LLMAgent #MolecularGeneration #75Properties #QED #Boltz2 #BindingAffinity #Scalable #Huynh #DrugDiscoveryToday #AIAgents #AutonomousReasoning #LiteratureSynthesis #ToxicityPrediction #MolecularDesign #EndToEnd #Zheng #Patterns #ParadigmShift #DiseaseMechanisms #DeNovo #ClinicalTrials #TargetDisease #ADMET #DeepeningDay #TechnicalDeepening #QuantitativeResults #BiomedicalAI #DailyReport*

---

### Post 4: 단백질 AI — AlphaFold Comprehensive Review × AlphaFold DB Updates: Deepening

🔬 **[단백질 AI] Malhotra et al.로 AlphaFold comprehensive review: AF1→AF2→AF3 evolution + drug discovery + vaccine design + dynamics limitations(Computers Biology Medicine 2025) + Fleming et al.로 AlphaFold DB + 3D-Beacons: AlphaMissense integration + Foldseek + variant pathogenicity + bulk download(NAR Genomics Bioinformatics 2025) + Kinde et al.로 AlphaFold disease applications: disease mechanisms + diagnostic markers + structural consequences + hybrid methods로 dynamics 극복(Results Engineering 2026): 단백질 AI Deepening(심화) 동시 확인**

단백질 AI의 **AlphaFold comprehensive review + DB updates + disease applications 심화(Deepening)**가 출처에서 동시에 확인된다[7][8][9].

**AlphaFold Comprehensive Review: AF1→AF2→AF3 Evolution + Drug Discovery + Vaccine Design[7]**
Yuktika Malhotra et al.(Computers in Biology and Medicine 2025)의 review는 AlphaFold 시리즈의 진화를 종합한다[7]. **AF1→AF2→AF3 evolution**: versions별 능력 발전. **AlphaFold revolution**: deep learning 기반 구조 예측 혁신. **Drug discovery applications**: 약물 개발 적용 확대. **Vaccine design**:vaccine 설계에서의 활용. **Protein dynamics limitations**: 단백질 동역학 한계. **Protein folding understanding**: 접힘 이해의 재정의. **Protein design + disease research**: 단백질 설계 및 질병 연구. **Experimental techniques integration**: 실험 기법과의 통합. **Hybrid methods**: hybrid 방법으로 dynamics 한계 극복 시도. **Future integration**: 향후 통합 전망. 핵심: AlphaFold comprehensive — 이행의 기술 체계 심화(Deepening).

**AlphaFold DB + 3D-Beacons: AlphaMissense + Foldseek + Variant Pathogenicity[8]**
Jennifer Fleming et al.(NAR Genomics Bioinformatics 2025)의 article은 AlphaFold DB의 최근 업데이트를 보고한다[8]. **AlphaMissense integration**: 인간 단백질 Missense 변이 병력성 예측. **Interactive heatmap**: 변이 데이터 residue-level 시각화. **pLDDT + pathogenicity scores toggle**: 구조 품질과 병력성 점수 전환 가능. **Foldseek integration**: 빠른 구조 검색 및 비교. **Bulk data download**: 대량 데이터 다운로드 옵션. **3D-Beacons enhanced**: 상세 annotation endpoints. **LevyLab homomeric models**: homomeric AlphaFold 2 모델. 핵심: AlphaFold DB — 이행의数据库 심화(Deepening).

**AlphaFold Disease/Vaccine/Design Applications: Dynamics 극복 위한 Hybrid Methods[9]**
(M. Kinde et al. Results in Engineering 2026 — Post 4에서 상세 기술). Disease mechanisms + diagnostic markers + structural consequences + hybrid methods로 dynamics 극복. 핵심: AlphaFold disease applications — 이행의 적용 심화(Deepening).

**10월 17일 Clinicalizing(Garcia AlphaFold zero-shot + Feldman adversarial + Mi protein design) vs 오늘의 차별점**: 10/17이 **Garcia Protein Science**으로 AlphaFold zero-shot de novo design moderate discrimination를, **Feldman Computational Structural**으로 AlphaFold3 adversarial 40% invariance를, **Mi Computational Structural**으로 protein design AI era review를 제시했다면, 오늘은 **Malhotra Computers Biology Medicine**으로 **AF1→AF2→AF3 evolution + drug discovery + vaccine design의 comprehensive review** 심화(Deepening)를, **Fleming NAR Genomics Bioinformatics**으로 **AlphaMissense + Foldseek + bulk download의 AlphaFold DB updates** 심화(Deepening)를, **Kinde Results Engineering**으로 **disease mechanisms + hybrid methods로 dynamics 극복의 적용 심화**(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → comprehensive review + DB updates + disease applications Deepening(10/21).

실무 함의: 단백질 AI 연구팀에서 Fleming et al.의 AlphaMissense integration을 variant pathogenicity analysis에 참조해야 한다.

*출처: [Malhotra AlphaFold Comprehensive Review Computers Biology Medicine 2025](https://consensus.app/papers/details/97539d9282a55cc7b207d41360287063/), [Fleming AlphaFold DB 3D-Beacons NAR Genomics Bioinformatics 2025](https://consensus.app/papers/details/72b6c54e89965cda92c8a4bc025888f7/), [Kinde AlphaFold Disease Applications Results Engineering 2026](https://consensus.app/papers/details/97539d9282a55cc7b207d41360287063/)*

*#ProteinAI #AlphaFold #Malhotra #ComputersBiologyMedicine #AF1 #AF2 #AF3 #Evolution #DrugDiscovery #VaccineDesign #Dynamics #ProteinFolding #Fleming #NARGenomics #Bioinformatics #AlphaFoldDB #AlphaMissense #Foldseek #VariantPathogenicity #BulkDownload #3DBeacons #Kinde #ResultsEngineering #DiseaseMechanisms #DiagnosticMarkers #HybridMethods #DeepeningDay #ComprehensiveReview #DatabaseUpdates #ApplicationDeepening #BiomedicalAI #DailyReport*

---

### Post 5: Single-Cell Multi-Omics Foundation Model — scGPT/Geneformer Zero-Shot Limitations + Perturbation Benchmark: Deepening

🧫 **[Single-Cell] Kedzierska et al.로 scGPT/Geneformer zero-shot limitations: 148 citations + reliability challenges + simpler methods outperform + discovery settings where labels unknown(Genome Biology 2025) + Csendes et al.로 perturbation benchmark: scGPT + scFoundation + mean baseline outperforms + biologically meaningful features outperform + low perturbation-specific variance(BMG Genomics 2025) + Yan et al.로 scGPT/Geneformer learnability: 2026 + task-dependent benefits + cell type annotation substantial + perturbation prediction limited + model size not guarantee + bigger not always better(BMG Genomics 2026): Single-Cell Deepening(심화) 동시 확인**

Single-cell foundation model의 **zero-shot limitations + perturbation benchmark 한계 인식 심화(Deepening)**가 출처에서 동시에 확인된다[10][11][12].

**scGPT/Geneformer Zero-Shot Limitations: Reliability Challenges + Simpler Methods Outperform[10]**
Kasia Kedzierska et al.(Genome Biology 2025)의 연구는 scGPT와 Geneformer의 zero-shot 성능을 엄격히 평가한다[10]. **Zero-shot evaluation**: 추가 훈련 없이 성능 평가. **Reliability challenges**: 실제应用中 신뢰성 문제. **Simpler methods outperform**: 단순한 방법이 더 나은 성능. **Discovery settings where labels unknown**: 레이블 없는 발견 설정. **Foundation model limitations**: foundation model의 실제 한계. **Rigorous evaluation importance**: 엄격한 평가의 중요성. **Geneformer + scGPT both**: 두 모델 모두 한계 확인. 핵심: scGPT/Geneformer zero-shot — 이행의 평가 심화(Deepening).

**Perturbation Benchmark: scGPT + scFoundation + Mean Baseline Outperforms[11]**
Gerold Csendes et al.(BMC Genomics 2025)의 연구는 single-cell foundation model의 perturbation prediction을 benchmark한다[11]. **scGPT + scFoundation benchmarked**: 두 모델 평가. **Mean baseline outperforms**:最简单的平均값 모델이 superior. **Biologically meaningful features outperform**: 생물학적으로 의미 있는 특징 활용 모델이 superior. **Perturbation-specific variance low**: Perturb-Seq benchmark dataset의 낮은 분산. **Suboptimal benchmark evaluation**: 현재 benchmark의 한계. **Model evaluation insights**: 모델 평가에 대한 중요한 통찰. 핵심: perturbation benchmark — 이행의 benchmark 한계 심화(Deepening).

**scGPT/Geneformer Learnability: Task-Dependent + Bigger Not Always Better[12]**
Yuhai Yan et al.(BMC Genomics 2026)의 연구는 single-cell LLM의 학습 가능성을 평가한다[12]. **Task-dependent benefits**: 태스크 의존적 이점. **Cell type annotation substantial gains**: 세포 유형 주석에서 상당한 이점. **Perturbation prediction limited gains**: perturbation 예측에서 제한적 이점. **Model size not guarantee improved performance**: 모델 크기 ≠ 성능 보장. **Bigger is not always better**: 대형 모델 함정. **Simple summary statistics**: 실제 perturbation 예측에서 단순 요약 통계 이상 파악 어려움. **Task-specific architectures need**: 태스크 특화 아키텍처 필요성. **Biologically-informed priors**: 생물학 정보 기반 priors 중요성. 핵심: scGPT/Geneformer learnability — 이행의 평가 프레임워크 심화(Deepening).

**10월 17일 Clinicalizing(Li perturbation benchmark + Dimitrov single-cell interpretation) vs 오늘의 차별점**: 10/17이 **Li Science Advances**으로 PerturbNet + 13 methods + 25 datasets + foundation models compress variance를, **Dimitrov Nature Reviews**으로 causal inference + mechanistic discovery를 제시했다면, 오늘은 **Kedzierska Genome Biology**으로 **scGPT/Geneformer zero-shot reliability challenges + simpler methods outperform**이라는 근본적 한계 인식(Deepening)을, **Csendes BMC Genomics**으로 **mean baseline outperforms + biologically meaningful features outperform + low perturbation-specific variance**라는 benchmark 체계 한계(Deepening)를, **Yan BMC Genomics**으로 **task-dependent + bigger not always better + biologically-informed priors need**라는 평가 프레임워크 심화(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → zero-shot reliability + benchmark limitations + evaluation framework Deepening(10/21).

실무 함의: Single-cell 연구팀에서 Kedzierska et al.의 zero-shot evaluation 결과를 foundation model deployment strategy에 반영해야 한다.

*출처: [Kedzierska scGPT Geneformer Zero-Shot Genome Biology 2025](https://consensus.app/papers/details/1574cad890b251c7af78919561e7d8d1/), [Csendes Perturbation Benchmark BMC Genomics 2025](https://consensus.app/papers/details/c900aeec4a385a3a88ba4bbaff68883f/), [Yan scGPT Geneformer Learnability BMC Genomics 2026](https://consensus.app/papers/details/db23619e10075c0eb0957368f25ba7de/)*

*#SingleCell #FoundationModel #scGPT #Geneformer #scFoundation #Kedzierska #GenomeBiology #ZeroShot #Reliability #SimplerMethods #Outperform #Csendes #BMCGenomics #PerturbationBenchmark #MeanBaseline #BiologicallyMeaningful #LowVariance #Yan #TaskDependent #BiggerNotBetter #TaskSpecific #BiologicallyInformed #DeepeningDay #FoundationModelLimitations #BenchmarkCritique #EvaluationFramework #BiomedicalAI #DailyReport*

---

### Post 6: FDA/규제 — CRISPR Approval Landscape + On/Off-Target Spectrum Deepening

🏛️ **[FDA/규제] Chakraborty et al.로 CRISPR therapeutics approval landscape: Casgevy UK/MHRA + FDA dual approval + exa-cel + lovo-cel + industry-academia collaboration + patent landscape(Molecular Therapy Nucleic Acids 2026) + Magdy et al.로 CRISPR on/off-target activities: exagamglogene + kayjayguran + PAM-altered + on-target/off-target spectrum(Trends Biotechnology 2026) + Chakraborty et al.로 CRISPR Cas9 clinical trials review: current clinical trials + therapy approval landscape + challenges + safety concerns + cost-effectiveness(Molecular Therapy Nucleic Acids 2026): FDA/규제 Deepening(심화) 동시 확인**

FDA/규제 영역의 **CRISPR approval landscape + on/off-target spectrum 심화(Deepening)**가 출처에서 동시에 확인된다[3][2][3重复].

**CRISPR Therapeutics Approval Landscape: Casgevy + Industry-Academia + Patent[3]**
(Chakraborty et al. Molecular Therapy. Nucleic Acids 2026 — Post 2에서 상세 기술). Casgevy UK/MHRA + FDA + exa-cel + lovo-cel + industry-academia collaboration + patent landscape. 핵심: CRISPR approval landscape — 이행의 규제 현실 심화(Deepening).

**CRISPR On/Off-Target Spectrum: exagamglogene + kayjayguran + PAM-Altered[2]**
(Magdy et al. Trends in Biotechnology 2026 — Post 2에서 상세 기술). exagamglogene autotemcel + kayjayguran abengcemeran + PAM-altered base editing + on-target/off-target spectrum. 핵심: CRISPR on/off-target — 이행의 분자적 규제 심화(Deepening).

**CRISPR Cas9 Clinical Trials: Challenges + Safety + Cost-Effectiveness[3]**
(Chakraborty et al. Molecular Therapy. Nucleic Acids 2026 — Post 2에서 상세 기술). Clinical trials + therapy approval + challenges + safety concerns + cost-effectiveness. 핵심: CRISPR clinical trials — 이행의 종합적 규제 과제 심화(Deepening).

**10월 17일 Clinicalizing(Shahzad FDA CGT 38 approvals + Cohn Lonvoguran NEJM + Sheikh Kresladi) vs 오늘의 차별점**: 10/17이 **Shahzad Clinical Pharmacology**으로 FDA CGT 38 approvals 86.8% orphan + 92.1% expedited를, **Cohn NEJM**으로 Lonvoguran Phase 3 RCT를, **Sheikh Annals**으로 Kresladi FDA approval을 제시했다면, 오늘은 **Chakraborty Molecular Therapy**으로 **Casgevy UK/MHRA/FDA dual approval + exa-cel + lovo-cel + industry-academia collaboration + patent landscape**의 종합적 규제 현실(Deepening)을, **Magdy Trends Biotechnology**으로 **exagamglogene + kayjayguran + PAM-altered의 molecular spectrum** 분자적 규제 심화(Deepening)를, **Chakraborty 재확인**으로 **challenges + safety concerns + cost-effectiveness의 종합적 과제**(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → approval landscape + molecular spectrum + comprehensive challenges Deepening(10/21).

실무 함의: Regulatory Affairs팀에서 Chakraborty et al.의 CRISPR approval landscape를 gene therapy regulatory strategy에 참조해야 한다.

*출처: [Chakraborty CRISPR Therapeutics Approval Landscape Molecular Therapy Nucleic Acids 2026](https://consensus.app/papers/details/68148f0062ec554ba4af1996dd3718cb/), [Magdy CRISPR On/Off-Target Activities Trends Biotechnology 2026](https://consensus.app/papers/details/2952c798ad0458e188c7cc56cffc4dec/)*

*#FDA #Regulation #Chakraborty #MolecularTherapy #NucleicAcids #Casgevy #UKMHRA #FDA #ExaCel #LovoCel #IndustryAcademia #PatentLandscape #Magdy #TrendsBiotechnology #OnTarget #OffTarget #Exagamglogene #Kayjayguran #PAMAltered #BaseEditing #MolecularSpectrum #Challenges #SafetyConcerns #CostEffectiveness #DeepeningDay #ApprovalReality #MolecularRegulation #ComprehensiveChallenges #BiomedicalAI #DailyReport*

---

### Post 7: Spatial Transcriptomics/Cancer Atlas — Pan-Cancer Spatial Microenvironment × TLS Atlas × SMTdb: Deepening

🗺️ **[Spatial] Li et al.로 pan-cancer spatial microenvironment: 373 samples + 12 cancer types + 56 LCPs + 13 niches + Niche_4 poor prognosis + Niche_11 better survival(Cell Reports Medicine 2026) + Cho et al.로 TLS pan-cancer atlas: 12 cancer types + TLS spatial architecture + maturation states + AI framework + H&E images + TCGA( Science 2026) + Zhou et al.로 SMTdb spatial meta-transcriptome: 203 slices + 20 cancer types + 334,253 spots + 1,908,646 cells + 1,218 microbiota + TME microbiome(SMTdb Molecular Biology Evolution 2025): Spatial Deepening(심화) 동시 확인**

Spatial transcriptomics/cancer atlas의 **pan-cancer spatial microenvironment + TLS atlas + SMTdb database 심화(Deepening)**가 출처에서 동시에 확인된다[13][14][15].

**Pan-Cancer Spatial Microenvironment: 373 Samples + 12 Cancer Types + 56 LCPs + 13 Niches[13]**
Jiarong Li et al.(Cell Reports Medicine 2026)의 연구는 12개 암종의 공간적 종양 미세환경을 분석한다[13]. **Pan-cancer spatial transcriptomic analysis**: 373개 샘플, 12개 암종. **56 local cellular programs(LCPs)**: 국소 세포 프로그램 56개. **13 recurrent niches**: 반복 niche 13개. **Niche_4 poor prognosis + immunotherapy resistance**: 종양세포+대식세포 colocalization → 불량 예후. **Niche_11 better survival + treatment response**: 면역세포 colocalization → 양호한 예후. **Ligand-receptor analysis**: niche-shared 및 niche-specific 상호작용. **Spatial organization drivers**: 공간 조직화 메커니즘. **Cell location-dependent gene expression**: 세포 위치 의존적 유전자 발현. 핵심: pan-cancer spatial microenvironment — 이행의 공간적 심화(Deepening).

**TLS Pan-Cancer Atlas: 12 Cancer Types + AI Framework + H&E Images[14]**
K. S. Cho et al.(Science 2026)의 연구는 tertiary lymphoid structure(TLS)의 pan-cancer atlas를 구축한다[14]. **Pan-cancer TLS atlas**: 12개 암종. **TLS spatial architecture + maturation states**: 공간적 구조 및 성숙 상태. **Coordinated niche cell population remodeling**: niche 세포 집합의 조정된 재모델링. **Distance-dependent gradients in tumor programs**: 종양 프로그램의 거리 의존적 구배. **AI framework for TLS maturation**: H&E 이미지로 TLS 성숙 상태 예측 AI. **TCGA + therapy cohorts validation**: 대규모 검증. **Maturation-aware composite score**: 성숙도 인식 복합 점수. **Conventional TLS metrics outperforms**: 기존 TLS 지표 superior. 핵심: TLS pan-cancer atlas — 이행의 공간적 구조 심화(Deepening).

**SMTdb: Spatial Meta-Transcriptome 203 Slices + 20 Cancer Types + 1,218 Microbiota[15]**
Wei-Wei Zhou et al.(Molecular Biology and Evolution 2025)의 연구는 종양 microbiome spatial database를 구축한다[15]. **SMTdb**: spatial meta-transcriptome database. **203 fresh frozen slices**: 20개 암종. **334,253 spots + 1,908,646 cells**: 대규모 공간 해상도. **1,218 microbiota in spatial tissue**: 공간 조직 내 미생물. **Tumor microenvironment(TME) microbiome**: 종양 미세환경 미생물군총. **Host gene modules + microbiota**: 숙주 유전자 모듈과 미생물 연관성. **Immune cell co-occurrence**: 면역세포 공존. **Comprehensive structured platform**: 포괄적 구조화 플랫폼. 핵심: SMTdb — 이행의 database 심화(Deepening).

**10월 17일 Clinicalizing(Zhang spatial ecotypes + Ji CAPTAIN multimodal) vs 오늘의 차별점**: 10/17이 **Zhang Nature**으로 spatial ecotypes 9 SEs + cfDNA deep learning를, **Ji Nature Communications**으로 CAPTAIN 4M+ cells + 382 proteins를 제시했다면, 오늘은 **Li Cell Reports Medicine**으로 **373 samples + 12 cancer types + 56 LCPs + 13 niches + Niche_4 poor prognosis + Niche_11 better survival**의 구체적 공간 미세환경 심화(Deepening)를, **Cho Science**으로 **TLS pan-cancer atlas + AI framework + H&E + TCGA validation**의 공간 구조 심화(Deepening)를, **Zhou Molecular Biology Evolution**으로 **SMTdb 203 slices + 20 cancer types + 1,218 microbiota + TME microbiome**의 database 심화(Deepening)를 동시에 제시한다. Clinicalizing(10/17) → spatial microenvironment + TLS atlas + SMTdb Deepening(10/21).

실무 함의: 암종 연구팀에서 Li et al.의 Niche_4/Niche_11 예후 연관성을 immunotherapy response prediction에 참조해야 한다.

*출처: [Li Pan-Cancer Spatial Microenvironment Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/), [Cho TLS Pan-Cancer Atlas Science 2026](https://consensus.app/papers/details/cb59d851e3205ee3acf590ee72f2a8c5/), [Zhou SMTdb Spatial Meta-Transcriptome Molecular Biology Evolution 2025](https://consensus.app/papers/details/61294565300e5c989faf97163c12000c/)*

*#Spatial #Transcriptomics #Li #CellReportsMedicine #PanCancer #SpatialMicroenvironment #373Samples #12CancerTypes #56LCPs #13Niches #Niche4 #Niche11 #PoorPrognosis #BetterSurvival #Cho #Science #TLS #PanCancerAtlas #SpatialArchitecture #MaturationStates #AICluster #HEV #TCGA #Zhou #MolecularBiology #Evolution #SMTdb #SpatialMetaTranscriptome #203Slices #20CancerTypes #334KSpots #1Point9MCells #1218Microbiota #TME #Microbiome #DeepeningDay #SpatialDeepening #DatabaseDeepening #BiomedicalAI #DailyReport*

---

### Post 8: 오늘의 요약 — Deepening Day

🧬 **AI × Bioresearch Daily — 10월 21일 요약: Deepening(심화)**

오늘 Deepening(심화)은 Clinicalizing(10/17)의 임상 결과들이 **실제 메커니즘 이해·검증 체계 확립·한계 인식으로 심화되는 날**이었다.

**CRISPR/Gene Editing — PE7 Methodology + On/Off-Target Spectrum + Approval Reality**[1][2][3]
Jong et al. BMB Reports 2026: PE1→PE7 evolution + "CRISPR 3.0" + in vivo delivery. Magdy et al. Trends Biotechnology 2026: exagamglogene + kayjayguran + PAM-altered base editing molecular spectrum. Chakraborty et al. Molecular Therapy 2026: Casgevy UK/MHRA/FDA + industry-academia + patent landscape. 핵심: CRISPR — 방법론 + 분자 스펙트럼 + 규제 현실의 심화.

**AI Drug Discovery — Modular Framework + AI Agents + Paradigm Shift**[4][5][6]
Ock et al. Journal Chemical Information 2025: LLM agent modular + 75 properties + QED >0.6 34→55 + Boltz-2 binding affinity. Huynh et al. Drug Discovery Today 2026: autonomous reasoning + literature synthesis + toxicity + end-to-end. Zheng et al. Patterns 2025: disease mechanisms + de novo + clinical trials paradigm shift. 핵심: AI drug discovery — 기술 체계 + 개념 + 패러다임의 심화.

**단백질 AI — AlphaFold Comprehensive + DB Updates + Disease Applications**[7][8][9]
Malhotra et al. Computers Biology Medicine 2025: AF1→AF2→AF3 evolution + drug discovery + vaccine design. Fleming et al. NAR Genomics Bioinformatics 2025: AlphaMissense + Foldseek + bulk download + variant pathogenicity. Kinde et al. Results Engineering 2026: disease mechanisms + hybrid methods + dynamics 극복. 핵심: 단백질 AI — comprehensive review + database + applications의 심화.

**Single-Cell Foundation Model — Zero-Shot Limitations + Benchmark Critique**[10][11][12]
Kedzierska et al. Genome Biology 2025: scGPT/Geneformer zero-shot reliability challenges + simpler methods outperform. Csendes et al. BMC Genomics 2025: mean baseline outperforms + low perturbation-specific variance. Yan et al. BMC Genomics 2026: task-dependent + bigger not always better + biologically-informed priors need. 핵심: Single-cell FM — zero-shot 한계 + benchmark 체계의 심화.

**FDA/규제 — Approval Landscape + Molecular Spectrum + Challenges**[3][2]
Chakraborty et al. Molecular Therapy 2026: Casgevy dual approval + exa-cel + lovo-cel + industry-academia + patent. Magdy et al. Trends Biotechnology 2026: on-target/off-target molecular spectrum. 핵심: FDA/규제 — 종합적 규제 현실의 심화.

**Spatial Transcriptomics — Pan-Cancer Microenvironment + TLS Atlas + SMTdb**[13][14][15]
Li et al. Cell Reports Medicine 2026: 373 samples + 12 cancers + 56 LCPs + 13 niches + Niche_4 poor/Niche_11 good prognosis. Cho et al. Science 2026: TLS pan-cancer atlas + AI framework + H&E + TCGA. Zhou et al. Molecular Biology Evolution 2025: SMTdb 203 slices + 20 cancers + 1,218 microbiota. 핵심: Spatial — microenvironment + TLS 구조 + database의 심화.

**전체 흐름**: Clinicalizing(10/17)된 Framework들이 Deepening(10/21)을 거쳐 **실제 메커니즘 심화·검증 체계 확립·한계 인식으로 나아가는 중**. 차주 Trend는 **Translation(전이)** — Deepening된 Framework들이 실제 치료 개발로 이전되는 단계 예상.

*출처: [Jong Prime Editing PE7 BMB Reports 2026](https://consensus.app/papers/details/11216802c06153dca45549e3d4ff98f5/), [Chakraborty CRISPR Therapeutics Landscape Molecular Therapy Nucleic Acids 2026](https://consensus.app/papers/details/68148f0062ec554ba4af1996dd3718cb/), [Magdy CRISPR On/Off-Target Trends Biotechnology 2026](https://consensus.app/papers/details/2952c798ad0458e188c7cc56cffc4dec/), [Ock LLM Agent Modular Journal Chemical Information Modeling 2025](https://consensus.app/papers/details/2332c070569c5b5385f063ecacff1003/), [Huynh AI Agents Drug Discovery Today 2026](https://consensus.app/papers/details/d47b0cf47a4f5055a7175a6fe09bb6e4/), [Zheng LLM Drug Discovery Patterns 2025](https://consensus.app/papers/details/5abbd25ca9d65110a7dacde7c5455b2e/), [Malhotra AlphaFold Review Computers Biology Medicine 2025](https://consensus.app/papers/details/97539d9282a55cc7b207d41360287063/), [Fleming AlphaFold DB NAR Genomics Bioinformatics 2025](https://consensus.app/papers/details/72b6c54e89965cda92c8a4bc025888f7/), [Kedzierska scGPT Geneformer Zero-Shot Genome Biology 2025](https://consensus.app/papers/details/1574cad890b251c7af78919561e7d8d1/), [Csendes Perturbation Benchmark BMC Genomics 2025](https://consensus.app/papers/details/c900aeec4a385a3a88ba4bbaff68883f/), [Yan scGPT Geneformer Learnability BMC Genomics 2026](https://consensus.app/papers/details/db23619e10075c0eb0957368f25ba7de/), [Li Pan-Cancer Spatial Microenvironment Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/), [Cho TLS Pan-Cancer Atlas Science 2026](https://consensus.app/papers/details/cb59d851e3205ee3acf590ee72f2a8c5/), [Zhou SMTdb Molecular Biology Evolution 2025](https://consensus.app/papers/details/61294565300e5c989faf97163c12000c/)*

*#DeepeningDay #Summary #PrimeEditing #PE7 #CRISPR3Point0 #OnTarget #OffTarget #PAMAltered #Casgevy #LLMAgent #ModularFramework #Boltz2 #AIAgents #AutonomousReasoning #AlphaFold #AlphaMissense #Foldseek #scGPT #Geneformer #ZeroShot #Perturbation #Spatial #PanCancer #TLSAtlas #SMTdb #Niche4 #Niche11 #Translation #ClinicalTransition #MechanisticUnderstanding #Validation #BiomedicalAI #DailyReport*

---

## Sources (1차 출처 14개 이상)

1. [Jong et al. — Prime editing PE7 review, BMB Reports 2026](https://consensus.app/papers/details/11216802c06153dca45549e3d4ff98f5/)
2. [Chakraborty et al. — CRISPR therapeutics approval landscape, Molecular Therapy. Nucleic Acids 2026](https://consensus.app/papers/details/68148f0062ec554ba4af1996dd3718cb/)
3. [Magdy et al. — CRISPR on/off-target activities, Trends in Biotechnology 2026](https://consensus.app/papers/details/2952c798ad0458e188c7cc56cffc4dec/)
4. [Ock et al. — LLM Agent modular framework, Journal of Chemical Information and Modeling 2025](https://consensus.app/papers/details/2332c070569c5b5385f063ecacff1003/)
5. [Huynh et al. — AI agents drug discovery, Drug Discovery Today 2026](https://consensus.app/papers/details/d47b0cf47a4f5055a7175a6fe09bb6e4/)
6. [Zheng et al. — LLM drug discovery review, Patterns 2025](https://consensus.app/papers/details/5abbd25ca9d65110a7dacde7c5455b2e/)
7. [Malhotra et al. — AlphaFold comprehensive review, Computers in Biology and Medicine 2025](https://consensus.app/papers/details/97539d9282a55cc7b207d41360287063/)
8. [Fleming et al. — AlphaFold DB + 3D-Beacons, NAR Genomics Bioinformatics 2025](https://consensus.app/papers/details/72b6c54e89965cda92c8a4bc025888f7/)
9. [Kedzierska et al. — scGPT/Geneformer zero-shot limitations, Genome Biology 2025](https://consensus.app/papers/details/1574cad890b251c7af78919561e7d8d1/)
10. [Csendes et al. — Perturbation benchmark, BMC Genomics 2025](https://consensus.app/papers/details/c900aeec4a385a3a88ba4bbaff68883f/)
11. [Yan et al. — scGPT/Geneformer learnability, BMC Genomics 2026](https://consensus.app/papers/details/db23619e10075c0eb0957368f25ba7de/)
12. [Li et al. — Pan-cancer spatial microenvironment, Cell Reports Medicine 2026](https://consensus.app/papers/details/f420ab985e1257e18352cac9e0c425cf/)
13. [Cho et al. — TLS pan-cancer atlas, Science 2026](https://consensus.app/papers/details/cb59d851e3205ee3acf590ee72f2a8c5/)
14. [Zhou et al. — SMTdb spatial meta-transcriptome, Molecular Biology and Evolution 2025](https://consensus.app/papers/details/61294565300e5c989faf97163c12000c/)
15. [Jihyeon Yu et al. — Evolution of Prime Editing, Advanced Science 2026](https://consensus.app/papers/details/475fc3cbbe805f5e915626f08d2da039/)