Research Pulse — 2026-08-21 | Brown Biotech

# Research Pulse — 2026-08-21
_Generated 2026-08-21 07:31 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 — 8월 21일 (금요일) · Translating(임상화 Day)**

8/21(임상화)→8/20(Expanding)→8/19(Diversifying)→8/18(Converging)→8/17(Standardizing)→8/16(Industrializing)→8/15(Sustaining)→8/14(Accelerating)→8/13(Catalyzing)→...→7/14(가속). Expanding(8/20)된 기술들이 **임상 적용/실제 환자 데이터로 Translating(임상화)하는 날** — VLP toolkit이 primary human myeloid cells(monocytes·macrophages·dendritic cells)에 CRISPR editing + pooled screening 전달(Jung Nature Biotechnology 2026), AlphaFold3이 TCR mimic antibody-pMHC 구조 예측으로 CAR therapy design 적용(Sarkar mAbs 2026), EAGLE deep learning이 whole-slide pathology image를 2.27초에 분석하여 computational 시간 99% 절감(Neidlinger Nature Communications 2026), CAPTAIN multimodal foundation model이 RNA+protein co-assay 4M cells로 학습하여 surface protein imputation 가능 확인(Ji Nature Communications 2026), FDA AI drug guidance가 risk-based credibility framework로 draft 단계 공개(Niazi Journal Chemistry 2026), kidney cancer cells가 embryonic nephrogenesis로 역행하여 ICI resistance mechanism 발견(Yarlagadda bioRxiv 2026), CRC adenoma에서 senescence-stemness spatial coupling이 immune exclusion驱动确认(Yu bioRxiv 2026)이 동시에 진행된다.

Translating은 Expanding의 다음 단계다. 오늘 6개 전선은 확산된 기술들이 **실제 환자 샘플·임상 데이터·실제 환자 적용으로 임상화하는 날**이다.

*출처: [Jung VLP Myeloid Cells Nature Biotechnology 2026](https://www.nature.com/articles/s41587-026-03258-2), [Sarkar AF3 TCRm mAbs 2026](https://doi.org/10.1080/19420862.2026.2719263), [Neidlinger EAGLE Nature Communications 2026](https://www.nature.com/articles/s41467-025-60225-4), [Ji CAPTAIN Nature Communications 2026](https://www.nature.com/articles/s41467-025-60225-4), [Niazi FDA AI Guidance Journal Chemistry 2026](https://consensus.app/papers/details/e016addd493055de8bbca57954a06666/), [Yarlagadda Kidney Cancer ICI Resistance bioRxiv 2026](https://doi.org/10.64898/2026.08.05.743137)*

*#TranslatingDay #VLP #Jung #NatureBiotechnology #MyeloidCells #CRISPR #Sarkar #mAbs #AlphaFold3 #TCRm #Neidlinger #EAGLE #Pathology #Ji #CAPTAIN #Multimodal #FoundationModel #Niazi #FDA #AIGuidance #Yarlagadda #KidneyCancer #Embryonic #ICIResistance #Yu #CRC #Senescence #Spatial #Translating #ClinicalTranslation #CAR Therapy #Immunotherapy #WholeSlideImage #DigitalPathology*

---

### Post 2: CRISPR & Gene Editing — VLP Toolkit Primary Myeloid Cells 적용과 Neuropsychiatric Prime Editing 임상화

🧪 **[CRISPR/Gene Editing] VLP toolkit이 primary human myeloid cells(monocytes·macrophages·dendritic cells)에 high-efficiency CRISPR editing + pooled CRISPR screening 전달(Jung Nature Biotechnology 2026) + Prime editing이 신경계 질환으로 임상 적용 시작 — blood-brain barrier penetration 전략 구체화(Lushington Molecular Therapy 2026 + Ji Neuroprotection 2026): CRISPR의 myeloid cell immunotherapy + 신경계 임상화 동시 확인**

CRISPR 기술의**임상 적용 확대**가 출처에서 동시에 확인된다[1][2][3].

**VLP Toolkit: Primary Myeloid Cells Targeted CRISPR Delivery[1]**
Hyuncheol Jung et al.(Nature Biotechnology 2026)의 연구는 VLP(virus-like particle) 기반 CRISPR toolkit을 제시한다. **대상 세포**: primary human monocytes, macrophages, dendritic cells — immunotherapy에서 중요한 세포 유형. **전달 효율**: 높은 viability와 innate immune responsiveness 유지하면서 high-efficiency delivery. **지원 modality**: gene knockout, base editing, epigenetic silencing 모두 가능. **SLICeVLP**: sgRNA는 VPX-lentivirus로, Cas9 protein은 engineered VLPs로 각각 전달하는 하이브리드 시스템. **Pooled screening 적용**: Perturb-seq screens로 human macrophages에서 TNF와 CD80 발현 조절자 발견. **TNFAIP3 동정**: inflammatory polarization의 central regulator로 확인 — proinflammatory state 유발, suppressive repolarization 저항. **CAR macrophages**: TNFAIP3 knockout이 enhanced cytotoxicity 유도. 핵심: VLP toolkit으로 primary myeloid cells targeted CRISPR screening + immunotherapy 적용 — innate immunity modulation의 임상 경로 확립.

**Prime Editing 임상화: Neuropsychiatric Disorders 적용[2]**
Caleb Lushington et al.(Molecular Therapy 2026)의 review는 prime editing의 first-in-human trial 결과를 분석한다. **첫 임상 결과**: functional restoration 확인 + promising safety profile 보고. **효율성 향상**: Cas variant selection, RT engineering, pegRNA improvements으로 지속 진전. **전달 진전**: nanoparticles, split viral systems로 질병 모델 적용 가속. **남은 도전**: efficiency, delivery, safety의 세 가지 과제. 핵심: PE first-in-human trial 완료 + functional restoration 확인 — prime editing의 임상화 공식 시작.

**Prime Editing 신경계 적용: BBB Penetration 전략 구체화[3]**
Tianshan Ji et al.(Neuroprotection 2026)의 review는 prime editing의 신경계 적용 장벽을 분석한다. **장벽들**: (1) post-mitotic neurons에서 낮은 editing efficiency, (2) 복잡한 pegRNA design, (3) reverse transcription errors, (4) vector payload limitations, (5) blood-brain barrier penetration. **진행 중인 해결책**: split-AAV systems, lipid nanoparticles, engineered peptides, compact Cas variants. **표적 질환**: monogenic neurodevelopmental disorders. **필요 조건**: improved editors, BBB-penetrant delivery, transient expression, genome-wide safety evaluation. 핵심: prime editing 신경계 적용의 장벽과 해결 전략 구체화 — 신경계 유전자 치료의 임상화 경로 명확화.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **Janjuha Nature Biomedical Engineering**으로 **ISS로 gene editing spatial profiling을 mice·macaques·종양/대사성으로 확산**을, **Froechlich Molecular Therapy**으로 **mRNA/LNP vaccine de-targeting framework**를, **Tálas Science Translational Medicine**으로 **RNA-LNP PE7로 CTLN1 치료 가능성**을 제시했다면, 오늘은 **Jung Nature Biotechnology**으로 **VLP toolkit로 primary myeloid cells에 CRISPR screening 적용**을(myeloid cell immunotherapy로), **Lushington Molecular Therapy**으로 **PE first-in-human trial 완료 + functional restoration**을(임상화 공식 시작), **Ji Neuroprotection**으로 **prime editing 신경계 적용 장벽과 BBB penetration 전략**을(신경계 임상화 경로) 동시에 제시한다. Expanding(8/20) → 임상 적용 Translating(8/21).

실무 함의: 면역세포 치료 개발팀에서 Jung et al.의 VLP toolkit 분석하여 myeloid cell immunotherapy pipeline 전략을 수립해야 한다.

*출처: [Jung VLP Myeloid Cells CRISPR Nature Biotechnology 2026](https://www.nature.com/articles/s41587-026-03258-2), [Lushington Prime Editing First-in-Human Molecular Therapy 2026](https://consensus.app/papers/details/79a173589b1d58598fd94fa334768881/), [Ji Prime Editing Neuropsychiatric Neuroprotection 2026](https://consensus.app/papers/details/6f2f6891edcd524b9e54932910f51d1d/)*

*#CRISPR #GeneEditing #Jung #NatureBiotechnology #VLP #MyeloidCells #Macrophages #DendriticCells #CAR #TNFAIP3 #Lushington #MolecularTherapy #PrimeEditing #FirstInHuman #Ji #Neuroprotection #BBB #PrimeEditing #Neuropsychiatric #TranslatingDay #InVivoEditing #ClinicalTrial #Immunotherapy #InnateImmunity #SplitAAV #LipidNanoparticle*

---

### Post 3: AI 신약 — EAGLE Pathology AI와 AlphaFold3 TCRm Antibody Design 적용 확산

💊 **[AI 신약] EAGLE deep learning이 whole-slide pathology image를 2.27초에 분석하여 computational 시간 99% 절감(Neidlinger Nature Communications 2026) + AlphaFold3이 TCR mimic antibody-pMHC 구조 예측으로 CAR therapy design 적용(Sarkar mAbs 2026) + AlphaFold mutation invariance 문제로 protein design 적용 한계 재확인(Feldman Computational Biotechnology 2026): AI pathology의 임상 적용 확산 + 단백질 AI의 한계 동시 확인**

AI-Driven Drug Discovery의**임상 적용 확산**이 출처에서 동시에 확인된다[4][5][6].

**EAGLE: Efficient Pathology AI — 2.27초 Whole-Slide Analysis[4]**
Patrick Neidlinger et al.(Nature Communications 2026)의 연구는 EAGLE deep learning framework를 제시한다. **문제**: 기존 whole-slide analysis가 thousands of redundant tiles 처리 + high-performance computing 필요. **EAGLE 혁신**: task-agnostic tile selection + detailed feature extraction로 pathologist 방식 emulation. **43개 task benchmark**: 9개癌種·morphology·biomarker prediction·treatment response·prognosis 포함. **성능**: patch aggregation 대비 up to 23% 향상, 전체 classification 최고 성능. **속도**: 1 slide를 2.27초에 처리 — 기존 대비 99% computational 시간 절감. **auditable workflow**: exact tiles 추적 가능 + attention concentration 분석으로 robust outputs. **적용**: rapid slide search, multi-omics pipelines integration, clinical foundation models. 핵심: EAGLE deep learning으로 whole-slide pathology AI 임상 적용 가능 — computational pathology의 real-time 분석 시대.

**AlphaFold3 TCR Mimic Antibody: CAR Therapy Design 적용[5]**
Abhijit Sarkar et al.(mAbs 2026)의 연구는 AlphaFold3의 TCR mimic(TCRm) antibody-pMHC 구조 예측을 분석한다. **AF3 예측 신뢰도**: TCRm Fv-pMHC structures에서 high confidence 달성. **NY-ESO-1/HLA-A*02:01 적용**: AF3 예측 intermolecular geometry가 canonical TCR engagement과 유사 확인. ** mutagenesis 검증**: predicted epitopes와 paratopes experimental 검증 완료. **AF3 aided design 실패**: larger Fv-pHLA contact surface + 더 많은 hydrogen bonds 설계했으나 binding avidity 향상 미확인. **결론**: AF3은 TCRm-pMHC complex modeling에 highly valuable + force field-based methods와 결합 필요. 핵심: AlphaFold3으로 TCRm antibody design 적용 확산 확인 + force field 보완 필요 — 구조 예측의 임상 적용 한계 명확화.

**AlphaFold Mutation Invariance: Protein Design 적용 한계[6]**
Jonathan Feldman et al.(Computational and Structural Biotechnology Journal 2026)의 연구는 AlphaFold의 adversarial mutation evaluation을 제시한다. **핵심 발견**: AlphaFold3가 point/deletion mutations에서 residues의 up to 40% invariance 보임 — including deliberately destabilizing substitutions. **fold-switching 단백질에서도 동일**: small monomeric 단백질에서도 known conformational changes 예측 실패. **confidence metrics 불신뢰**: most accurate structure을 35%에서만 select. **ESMFold 비교**: 더 나은 mutational sensitivity 보이나 여전히 imperfect. **함의**: AlphaFold가 biophysical reasoning보다 template-based pattern matching에 의존 — mutation-effect interpretation, confidence-guided model selection, sequence optimization workflow에 직접적 영향. 핵심: AlphaFold의 mutation invariance 문제再確認 — protein design applications의 근본적 한계 제시.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **Zhang Drug Development Research**으로 **DrugCLIP contrastive learning genome-wide screening**을, **Raghupathi Health Information Science Systems**으로 **4Vs framework 확장**을, **de Oliveira Teixeira Journal Computer-Aided Molecular Design**으로 **AlphaFold antibiotic resistance 적용**을 제시했다면, 오늘은 **Neidlinger Nature Communications**으로 **EAGLE pathology AI 99% computational 절감 + 임상 적용**을(실제 진단 적용), **Sarkar mAbs**으로 **AlphaFold3 TCRm CAR therapy design 적용 + 한계**를(구조 예측 임상 적용), **Feldman Computational Biotechnology**으로 **AlphaFold mutation invariance 문제再確認**을(단백질 design 한계) 동시에 제시한다. Expanding(8/20) → 임상 적용 Translating(8/21).

실무 함의: 병리 AI팀에서 Neidlinger et al.의 EAGLE framework 분석하여 clinical pathology AI deployment strategy를 수립해야 한다.

*출처: [Neidlinger EAGLE Pathology Nature Communications 2026](https://www.nature.com/articles/s41467-025-60225-4), [Sarkar AF3 TCRm mAbs 2026](https://doi.org/10.1080/19420862.2026.2719263), [Feldman AlphaFold Mutation Invariance Computational Biotechnology 2026](https://consensus.app/papers/details/68f35194db525fb68a7bc8c10546001b/)*

*#AI #DrugDiscovery #Neidlinger #NatureCommunications #EAGLE #PathologyAI #WholeSlideImage #Sarkar #mAbs #AlphaFold3 #TCRm #CAR #Antibody #Feldman #AlphaFold #MutationInvariance #ProteinDesign #TranslatingDay #DeepLearning #ComputationalPathology #ClinicalAI #StructuralBiology #BiophysicalReasoning*

---

### Post 4: 단백질 AI — CAPTAIN Multimodal Foundation Model과 G6PC1 Variant Classification 적용 확산

🔬 **[단백질 AI] CAPTAIN multimodal foundation model이 RNA+protein co-assay 4M cells로 학습하여 surface protein imputation 가능 확인(Ji Nature Communications 2026) + AlphaMissense로 GSD1a G6PC1 variants 임상 분류 체계 확립(Stein bioRxiv 2026) + Foundation model perturbation prediction의 attention mechanism이 gene-level features 이상의 added value 없음 확인(Kendiukhov BMC Genomics 2026): multimodal 단백질 AI의 임상 적용 한계 동시 확인**

단백질 AI의**임상 적용 확대와 한계**가 출처에서 동시에 확인된다[7][8][9].

**CAPTAIN: Multimodal RNA+Protein Foundation Model[7]**
Boya Ji et al.(Nature Communications 2026)의 연구는 CAPTAIN foundation model을 제시한다. **기존 한계 극복**: 기존 single-cell foundation models가 transcriptomes만 사용 → biased and partial cellular state characterization. **CAPTAIN 혁신**: 4M+ cells에서 concurrently measured transcriptomes + 382 surface proteins로 학습. **unified multimodal representations**: cross-modality dependencies modeling + cellular states diversity capture. **하류 task 성능**: protein imputation and expansion, cell type annotation, batch harmonization 모두 superior. **COVID-19 발견**: immune interaction patterns 관련 potential findings. **제로샷 성능**: robust generalization 확인. 핵심: CAPTAIN multimodal foundation model로 RNA+protein 통합 분석 가능 — surface protein imputation의 임상적 가치 제시.

**AlphaMissense G6PC1: Glycogen Storage Disease Type 1a Variant Classification[8]**
Richard Stein et al.(bioRxiv 2026)의 연구는 AlphaMissense의 GSD1a variant classification 적용을 분석한다. **질병**: glycogen storage disease type 1a(GSD1a) — G6PC1 missense variants. **AlphaMissense 활용**: 78개 missense variants의 pathogenicity prediction. **핵심 발견**: folded G6PC1 abundance와 catalytic capacity의 강한 선형 관계 — 대부분의 variants에서. **pathogenic mechanism**: compromised stability + unfolded protein response activation. **allosteric outliers**: relatively high abundance yet low activity cluster — active site 인접 sidechain network. **치료 표적**: molecular phenotype 기반 clinical (re)classification + therapeutic design directions 제시. 핵심: AlphaMissense로 GSD1a variant致病机理 분석 + 임상 분류 체계 확립 — AI-driven variant classification의 임상 적용.

**Foundation Model Attention: Gene-Level Features 이상의 Added Value 없음[9]**
Ihor Kendiukhov et al.(BMC Genomics 2026)의 연구는 single-cell foundation model의 attention interpretability를 체계적으로 평가한다. **평가 프레임워크**: 37개 분석 + 153개 statistical tests — scGPT, Geneformer V2-316M 포함. **Objective A(GRN recovery)**: attention patterns이 biologically structured 정보 encoding — CSSI로 up to [Formula] 향상. **Objective B(perturbation-target prediction)**: gene-level baselines(AUROC 0.81-0.88)가 attention edges(AUROC 0.70)보다 superior. **augmentation 효과**: gene-level predictors에 attention edges 추가 시 ΔAUROC = ~0 — incremental value 없음. **causal ablation**: attention heads knockout해도 degradation 없음. **권장**: perturbation prediction에서 trivial baseline 먼저 적용 + incremental value test 필수. 핵심: foundation model attention mechanism이 perturbation prediction에서 gene-level features 이상의 added value 없음 — interpretability 한계 명확화.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **de Oliveira Teixeira Journal Computer-Aided Molecular Design**으로 **AlphaFold antibiotic resistance 적용**을, **Tortolani Biochemical Pharmacology**으로 **GPR18 AlphaFold modeling neuroinflammatory disorders**를, **Dhir RSC Advances**으로 **AI drug design workflow 포괄적 종합**을 제시했다면, 오늘은 **Ji Nature Communications**으로 **CAPTAIN multimodal RNA+protein model로 surface protein imputation**을( multimodal 확장), **Stein bioRxiv**으로 **AlphaMissense GSD1a variant classification 임상 적용**을(단백질 AI 임상 적용), **Kendiukhov BMC Genomics**으로 **foundation model attention mechanism 한계再確認**을(interpretability 한계) 동시에 제시한다. Expanding(8/20) → 임상 적용 + 한계 동시 확인 Translating(8/21).

실무 함의: 전산단백질팀에서 Kendiukhov et al.의 평가 프레임워크 참조하여 foundation model perturbation prediction 신뢰성 assessment protocol을 수립해야 한다.

*출처: [Ji CAPTAIN Multimodal Nature Communications 2026](https://www.nature.com/articles/s41467-025-60225-4), [Stein G6PC1 AlphaMissense bioRxiv 2026](https://doi.org/10.64898/2026.07.27.741017), [Kendiukhov Foundation Model Attention BMC Genomics 2026](https://consensus.app/papers/details/a35e627954f2569dad1c81b6f3d42f60/)*

*#ProteinAI #CAPTAIN #Multimodal #NatureCommunications #RNA #Protein #FoundationModel #Stein #AlphaMissense #GSD1a #VariantClassification #GlycogenStorageDisease #Kendiukhov #BMCGenomics #Attention #PerturbationPrediction #Geneformer #scGPT #TranslatingDay #CellState #ProteinImputation #BatchHarmonization #GRN #Interpretability #FoundationModel*

---

### Post 5: Single-Cell Multi-Omics Foundation Model — Learnability 평가와 scRNA-seq-ADT Integration 적용

🬬 **[Single-Cell Multi-Omics Foundation Model] Foundation model의 learnability과 scaling laws 체계적 평가로 perturbation prediction에서 한계 확인(Yan BMC Genomics 2026) + CEACAM6/KRT19 dual regulators가 CRC tumor heterogeneity와 mortality drivers 확인(Jing FASEB Journal 2026): single-cell multi-omics의 임상 적용 확인 동시**

Single-Cell Multi-Omics의**임상 적용 확대**가 출처에서 동시에 확인된다[10][11].

**Foundation Model Learnability: Perturbation Prediction에서 Scaling Laws 한계[10]**
Yuhai Yan et al.(BMC Genomics 2026)의 연구는 single-cell foundation models(scFMs)의 learnability를 체계적으로 평가한다. **평가 대상**: Geneformer, scGPT — perturbation prediction과 cell type annotation tasks. **핵심 발견 #1**: large-scale pretraining의 이점이 task-dependent — cell type annotation에서는 substantial advantages, perturbation prediction에서는 limited gains. **핵심 발견 #2**: model size 증가가 performance 보장 못함 — 오히려 detrimental 가능. **실험적 통찰**: perturbation prediction에서 tested scFMs가 simple summary statistics 이상의 것을 capture 못할 가능성. **권장**: task-specific architectures + biologically-informed priors으로 진전 필요. 핵심: foundation model의 scaling laws 한계再確認 — perturbation prediction의 근본적 문제 제시.

**CEACAM6/KRT19: CRC Tumor Heterogeneity Dual Regulators[11]**
Changwen Jing et al.(FASEB Journal 2026)의 연구는 CRC organoid-based multi-omics로 heterogeneity drivers를 동정한다. **방법**: 17개 patient-derived CRC organoids(85% success rate) + high-throughput drug screening + single-cell transcriptomics. **극단적 표현형 선별**: irinotecan+raltitrexed IC50 차이로 sensitive/resistant organoids 선별. **핵심 발견**: cluster_0 — REG4/TFF1/TFF3 및 epithelial keratins(KRT19/KRT8/KRT18) overexpression의 epithelial subpopulation expansion. **ITH score**: intratumoral heterogeneity score가 prognostic determinant로 확립 — stemness/angiogenesis metrics 대비 superior. **dual regulators**: CEACAM6와 KRT19이 mortality와 ITH의 dual regulators로 확인 — epithelial plasticity networks 통해 작용. **32개 ligand-receptor pairs**: intra-epithelial communication resistance hub mediation. 핵심: CEACAM6/KRT19 dual regulators로 CRC tumor heterogeneity mechanisms 분석 — precision oncology 적용.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **Kendiukhov Computational Biology Chemistry**으로 **foundation model adversarial validation framework**를, **Abir NAR Genomics Bioinformatics**으로 **CFM-GP conditional flow matching**를, **Fan Synthetic Systems Biotechnology**으로 **scYeast biological-knowledge-guided model**을 제시했다면, 오늘은 **Yan BMC Genomics**으로 **foundation model learnability 평가로 perturbation prediction 한계**를(scaling laws 한계), **Jing FASEB Journal**으로 **CEACAM6/KRT19 dual regulators로 CRC heterogeneity 분석**을(실제 환자 organoid 적용) 동시에 제시한다. Expanding(8/20) → 임상 데이터 적용 Translating(8/21).

실무 함의: 단일세포 연구팀에서 Jing et al.의 patient-derived organoid 접근법을 분석하여 CRC precision oncology strategy를 수립해야 한다.

*출처: [Yan Foundation Model Learnability BMC Genomics 2026](https://consensus.app/papers/details/db23619e10075c0eb0957368f25ba7de/), [Jing CEACAM6 KRT19 CRC FASEB Journal 2026](https://doi.org/10.1096/fj.202600610R)*

*#SingleCell #MultiOmics #Yan #BMCGenomics #FoundationModel #Learnability #ScalingLaws #Geneformer #scGPT #Jing #FASEBJournal #CEACAM6 #KRT19 #CRC #Organoid #ITH #Heterogeneity #PrecisionOncology #DrugResistance #TranslatingDay #CellTypeAnnotation #PerturbationPrediction #PatientDerived*

---

### Post 6: FDA/규제 — FDA AI Guidance Draft와 Exosome Therapeutic 임상화 현황

📋 **[FDA/규제] FDA AI drug guidance가 risk-based credibility framework로 draft 공개 — discovery/operation 단계 미포함 문제 확인(Niazi Journal Chemistry 2026) + Exosome therapeutic 임상화 현황 — FDA 승인 0건 + GMP manufacturing 진전 + engineered exosomes Phase I-IIb 완료(Shkurnikov Current Medicinal Chemistry 2026): FDA 규제 framework의 임상 적용 한계와 실제 동시 확인**

FDA/규제의**임상 적용 현황**이 출처에서 동시에 확인된다[12][13].

**FDA AI Guidance Draft: Risk-Based Credibility Framework[12]**
Sarfaraz K. Niazi(Journal Chemistry 2026)의 review는 FDA의 AI drug guidance를 비판적으로 분석한다. **FDA draft guidance(January 2025)**: AI의 regulatory decision-making 적용 위한 inaugural guidance. **강점**: structured risk-based credibility framework 제시. **문제점 #1**: scope이 regulatory decision-making으로 한정 — discovery와 operation phases 미포함. **문제점 #2**: bias mitigation strategies 강화 필요. **문제점 #3**: explainability requirements tiered system 부재. **문제점 #4**: real-world evidence integration 미흡. **문제점 #5**: EMA, NMPA, PMDA와의 global harmonization 부족. **권장사항**: 전체 lifecycle 포함 확장, bias mitigation 강화, tiered explainability, real-world evidence, international harmonization. 핵심: FDA AI drug guidance draft의 strengths와 한계 동시 분석 — AI pharmaceutical regulation의 현재 상태 제시.

**Surgical Robot FDA Recall: 20년간 281개 제품 분석[14]**
Rafał B. Drobot et al.(Journal Robotic Surgery 2026)의 연구는 FDA recall records 분석으로 surgical robot failure modes를 체계적으로 분류한다. **대상**: 2005-2026년 FDA Medical Device Recall Database 기준 281개 제품 중 274개 eligible. **failure modes 분류**: mechanical/material integrity가 40%로 가장 높고, labeling/IFU/regulatory 13.3%, software/control logic 12.5%, electrical/electronic/power 10.8% 순. **failure loci**: core system이 41.7%, instrument/accessory가 36.7%. **cybersecurity**: Versius Secure Boot가 유일한 explicit cybersecurity mechanism recall. **median initiation-to-posting interval**: 33일(IQR 22-111일). **함의**: surgical robot recalls가 mechanical failures이 주도 — software보다 hardware 문제가 더 빈번. 핵심: FDA recall data로 surgical robot failure modes 체계적 분류 — medical device post-market surveillance의 실제.

**Exosome Therapeutic: 임상화 현황과 규제 장벽[13]**
Maxim Shkurnikov et al.(Current Medicinal Chemistry 2026)의 review는 therapeutic exosomes의 임상화 현황을 종합한다. **현황(2025-2026)**: FDA, EMA 또는 동등 기관으로부터 승인된 extracellular vesicle 치료제는 0건. **완료된 trials**: Phase I-IIb trials — oncology와 pulmonology에서 favorable safety profiles 확인. **주요 도전**: (1) validated potency assays 부재, (2) batch consistency 문제, (3) regulatory harmonisation 미흡. ** manufacturing 진전**: mesenchymal stromal cell-derived exosomes dominance, scalable 3D hollow-fiber bioreactor로 GMP-grade manufacturing 가능. **순환 반감기**: cell line-derived EVs 2-30분, platelet-derived EVs 5.3-5.8시간. **CD47 + PEGylation**: 반감기 연장 가능. **AI-driven manufacturing optimisation**: 미래 전략으로 제시. 핵심: Exosome therapeutic 임상화 현황 — 승인 0건 + manufacturing 진전 + 규제 harmonisation 필요.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **Staton Cancer Biology Therapy**으로 **CAPItello-281 biomarker-guided FDA approval**을, **Morimoto Regenerative Therapy**으로 **ALS three converging approaches**를, **Del Toro Runzer Bioactive Materials**으로 **BMP-7 mRNA 골 재생**을 제시했다면, 오늘은 **Niazi Journal Chemistry**으로 **FDA AI guidance draft의 strengths와 한계**를(규제 framework 분석), **Drobot Journal Robotic Surgery**으로 **surgical robot FDA recall failure modes 체계적 분류**를(post-market surveillance), **Shkurnikov Current Medicinal Chemistry**으로 **exosome therapeutic 임상화 현황 + 규제 장벽**을(실제 임상 상태) 동시에 제시한다. Expanding(8/20) → 규제 실제 + 한계 동시 확인 Translating(8/21).

실무 함의: 규제 기획팀에서 Niazi et al.의 FDA AI guidance 분석을 참조하여 AI pharmaceutical regulatory strategy를 수립해야 한다.

*출처: [Niazi FDA AI Guidance Journal Chemistry 2026](https://consensus.app/papers/details/e016addd493055de8bbca57954a06666/), [Drobot Surgical Robot FDA Recall Journal Robotic Surgery 2026](https://doi.org/10.1007/s11701-026-03818-3), [Shkurnikov Exosome Therapeutic Current Medicinal Chemistry 2026](https://consensus.app/papers/details/cb59705f23e85fc5b6d424ac634a7fc8/)*

*#FDA #Regulation #Niazi #JournalChemistry #AIGuidance #CredibilityFramework #Drobot #SurgicalRobot #FDArecall #FailureModes #Shkurnikov #Exosome #ExtracellularVesicle #ClinicalTrials #GMP #Manufacturing #TranslatingDay #RiskBased #Bias #Explainability #MedicalDevice #PostMarketSurveillance*

---

### Post 7: Spatial Transcriptomics — Embryonic Nephrogenesis ICI Resistance와 Senescence-Stemness Spatial Coupling

🗺️ **[Spatial Transcriptomics] Kidney cancer cells가 embryonic nephrogenesis로 역행하여 ICI resistance mechanism 발견(Yarlagadda bioRxiv 2026) + CRC adenoma-carcinoma progression에서 senescence-stemness spatial coupling이 immune exclusion驱动(Yu bioRxiv 2026) + PDAC perineural niche spatial atlas로 nerve invasion mechanism 분석(Hussain bioRxiv 2026): spatial transcriptomics의 임상 resistance mechanism 해석 동시**

Spatial Transcriptomics의**임상 resistance mechanism 해석**이 출처에서 동시에 확인된다[15][16][17].

**Kidney Cancer Embryonic Reversion: ICI Resistance Mechanism[15]**
Dig Vijay Kumar Yarlagadda et al.(bioRxiv 2026)의 연구는 ccRCC의 ICI resistance mechanism을 spatial transcriptomics로 분석한다. **방법**: 110개 ccRCC 환자에서 single-cell RNA-seq + imaging-based spatial transcriptomics 결합. **핵심 발견**: cancer cells가 embryonic nephrogenesis trajectory( nephron progenitor → pretubular aggregate → renal vesicle → S-shaped body)로 연속적으로 분포. **ICI enrichment**: immune checkpoint inhibitors가 nephrogenic developmental trajectory로 cancer cells를 enrichment. **Notch signaling up-regulation**: persister cells에서 nephrogenic program + injury repair programs 동시 상승. **inhibitory immune checkpoint ligands**: 함께 up-regulation. **mouse model 검증**: immunocompetent ccRCC model에서 획득적 항-CTLA-4 resistance 확인. ** combiner target**: nephrogenic developmental program + inhibitory checkpoint repertoire. 핵심: kidney cancer cells가 embryonic nephrogenesis로 역행하여 ICI resistance mechanism发动 — developmental reversion의 치료 표적화.

**CRC Adenoma: Senescence-Stemness Spatial Coupling[16]**
Ming Yu et al.(bioRxiv 2026)의 연구는 CRC adenoma-carcinoma progression에서 senescence의 역할을 spatial multi-omics로 분석한다. **방법**: 24개 adenomas에서 Visium CytAssist + protein co-detection + Xenium Prime 5K profiling(101 patient-matched cores). **9개 spatial clusters 동정**: 두 dysplastic epithelial populations이 stemness, proliferation, senescence programs 동시 co-expression. **공간적 niche**: shared, spatially confined epithelial niche가 adenoma에서 carcinoma로 expansion. **GDF15 역할**: senescence-associated secretory factor로 senescence-stemness coupling mediation. **immune exclusion**: GDF15-high epithelium이 CD8+ T cells를 locally exclude — adenoma와 carcinoma 모두에서. **세포死亡驱动**: senescence가 tumor-suppressive가 아닌 spatially instructive program으로 기능. 핵심: senescence-stemness spatial coupling으로 CRC immune exclusion mechanism 해석 — cancer prevention 표적 제시.

**PDAC Perineural Niche: Spatial Atlas[17]**
Zainab Hussain et al.(bioRxiv 2026)의 연구는 PDAC perineural niche의 spatial atlas를 제시한다. **방법**: imaging-based spatial transcriptomics + pathology-guided single-nucleus RNA-seq 결합. **invaded nerve neighborhoods**: classical subtype cancer cells + myofibroblastic CAFs + lipid-associated macrophages. **non-invaded nerve neighborhoods**: inflammatory CAFs + B and T lymphocyte infiltration. **핵심 통찰**: PNI(past-neural invasion) 자체가 immune-suppressive — innervation 자체가 아니라. **AI 모델**: vision-centric deep learning model로 imaging-based spatial transcriptomics 수행. 핵심: PDAC perineural niche spatial atlas로 nerve invasion mechanism 분석 — PNI의 immune suppressive 역할 명확화.

**8월 20일 Expanding Day vs 오늘의 차별점**: 8/20이 **Koppensteiner Oncoimmunology**으로 **CAF-T cell adenosine axis로 NSCLC 예후 biomarker**을, **Lang Cancer Biology Therapy**으로 **CD3+ B cells의 DLBCL microenvironment 재구성**을, **Walberg Gut Microbes**으로 **bacterial LPS-CRC immune crosstalk**을 제시했다면, 오늘은 **Yarlagadda bioRxiv**으로 **embryonic nephrogenesis로 ICI resistance mechanism**을(developmental reversion), **Yu bioRxiv**으로 **senescence-stemness spatial coupling으로 immune exclusion**을(adenoma-carcinoma progression), **Hussain bioRxiv**으로 **PDAC perineural niche spatial atlas로 PNI mechanism**을(nerve invasion) 동시에 제시한다. Expanding(8/20) → resistance mechanism spatial 해석 Translating(8/21).

실무 함의: 면역항암 연구팀에서 Yarlagadda et al.의 embryonic reversion 분석을 참조하여 ICI combination therapy strategy를 수립해야 한다.

*출처: [Yarlagadda Kidney Cancer Embryonic Nephrogenesis ICI Resistance bioRxiv 2026](https://doi.org/10.64898/2026.08.05.743137), [Yu CRC Senescence Stemness Spatial bioRxiv 2026](https://doi.org/10.64898/2026.08.07.743310), [Hussain PDAC Perineural Niche bioRxiv 2026](https://doi.org/10.64898/2026.08.05.743048)*

*#SpatialTranscriptomics #Yarlagadda #KidneyCancer #ccRCC #Nephrogenesis #ICIResistance #Embryonic #Yu #CRC #Senescence #Stemness #GDF15 #ImmuneExclusion #Adenoma #Carcinoma #Hussain #PDAC #Perineural #NerveInvasion #CAF #Macrophages #TranslatingDay #SpatialOmics #ImmuneCheckpoint #CancerAtlas #Notch #DevelopmentalReversion*

---

### Post 8: 오늘의 요약 — Translating(임상화 Day) 핵심 정리

🧬 **오늘 6개 영역 핵심 요약:**

8/21(Translating)의 핵심: Expanding(8/20)된 기술들이 **임상 적용/실제 환자 데이터로 Translating(임상화)하는 날**이었다.

**1. VLP Toolkit Myeloid Cell Immunotherapy(Jung Nature Biotechnology 2026)**: Primary human myeloid cells(monocytes·macrophages·dendritic cells)에 VLP CRISPR toolkit로 gene knockout·base editing·epigenetic silencing 전달 + pooled Perturb-seq screening 적용 — TNFAIP3 central regulator 발견. [Jung Nature Biotechnology 2026]

**2. Prime Editing First-in-Human Trial 완료(Lushington Molecular Therapy 2026)**: PE first-in-human trial에서 functional restoration 확인 + promising safety profile — 신경계 적용은 BBB penetration 전략 구체화 중(Ji Neuroprotection 2026). [Lushington Molecular Therapy 2026]

**3. EAGLE Pathology AI 99% Computational 절감(Neidlinger Nature Communications 2026)**: Whole-slide pathology image를 2.27초에 분석 — computational 시간 99% 절감 + 43 tasks benchmark에서 23% 성능 향상. [Neidlinger Nature Communications 2026]

**4. CAPTAIN Multimodal RNA+Protein Model(Ji Nature Communications 2026)**: 4M cells + 382 surface proteins로 학습한 multimodal foundation model — surface protein imputation과 cell type annotation에서 superior performance. [Ji Nature Communications 2026]

**5. FDA AI Guidance Draft 분석(Niazi Journal Chemistry 2026)**: Risk-based credibility framework 강점 + discovery/operation phases 미포함, bias mitigation, explainability tiered system 부족 등 한계 동시 확인. [Niazi Journal Chemistry 2026]

**6. Kidney Cancer Embryonic Nephrogenesis ICI Resistance(Yarlagadda bioRxiv 2026)**: Cancer cells가 embryonic nephrogenesis로 역행하여 ICI resistance发动 — nephrogenic program + inhibitory checkpoint combination 표적 제시. [Yarlagadda bioRxiv 2026]

---

### Sources (1차 출처 10개 이상 URL)

1. https://www.nature.com/articles/s41587-026-03258-2 — Jung VLP Myeloid Cells CRISPR Nature Biotechnology 2026
2. https://consensus.app/papers/details/79a173589b1d58598fd94fa334768881/ — Lushington Prime Editing First-in-Human Molecular Therapy 2026
3. https://consensus.app/papers/details/6f2f6891edcd524b9e54932910f51d1d/ — Ji Prime Editing Neuropsychiatric Neuroprotection 2026
4. https://www.nature.com/articles/s41467-025-60225-4 — Neidlinger EAGLE Pathology Nature Communications 2026
5. https://doi.org/10.1080/19420862.2026.2719263 — Sarkar AF3 TCRm mAbs 2026
6. https://consensus.app/papers/details/68f35194db525fb68a7bc8c10546001b/ — Feldman AlphaFold Mutation Invariance Computational Biotechnology 2026
7. https://www.nature.com/articles/s41467-025-60225-4 — Ji CAPTAIN Multimodal Nature Communications 2026
8. https://doi.org/10.64898/2026.07.27.741017 — Stein G6PC1 AlphaMissense bioRxiv 2026
9. https://consensus.app/papers/details/a35e627954f2569dad1c81b6f3d42f60/ — Kendiukhov Foundation Model Attention BMC Genomics 2026
10. https://consensus.app/papers/details/db23619e10075c0eb0957368f25ba7de/ — Yan Foundation Model Learnability BMC Genomics 2026
11. https://doi.org/10.1096/fj.202600610R — Jing CEACAM6 KRT19 CRC FASEB Journal 2026
12. https://consensus.app/papers/details/e016addd493055de8bbca57954a06666/ — Niazi FDA AI Guidance Journal Chemistry 2026
13. https://doi.org/10.1007/s11701-026-03818-3 — Drobot Surgical Robot FDA Recall Journal Robotic Surgery 2026
14. https://consensus.app/papers/details/cb59705f23e85fc5b6d424ac634a7fc8/ — Shkurnikov Exosome Therapeutic Current Medicinal Chemistry 2026
15. https://doi.org/10.64898/2026.08.05.743137 — Yarlagadda Kidney Cancer Embryonic Nephrogenesis bioRxiv 2026
16. https://doi.org/10.64898/2026.08.07.743310 — Yu CRC Senescence Stemness Spatial bioRxiv 2026
17. https://doi.org/10.64898/2026.08.05.743048 — Hussain PDAC Perineural Niche bioRxiv 2026
18. https://doi.org/10.1016/j.mocell.2026.100391 — Woo CRISPR-dCas9-TET1 Liver Cancer Epigenetic Therapy Molecular Cells 2026
19. https://doi.org/10.64898/2026.08.04.742864 — Dohmen AlphaFold Limitation FT-IR bioRxiv 2026
20. https://doi.org/10.1007/s42770-026-02049-w — Dong Single-Cell Brucella Vaccine Candidate Brazilian Journal Microbiology 2026