CAR-T Cell Therapy
✓ 已验证Education · Service Delivery · Technical Validation · Three-in-One
Chimeric Antigen Receptor T-cell Therapy--teaching patients' own immune cells to precisely hunt down tumors. This page serves three audiences: patients and public wanting to understand CAR-T, CAR-T teams seeking computational service partners, and investors and peers evaluating technical capabilities.
Read "What is CAR-T", "Workflow", "Milestones", "Current Challenges" below--change your understanding of the therapy.
Focus on "DiVo's Role", "5-Step Pipeline", "Verified Capabilities"--we have sample reports available.
Focus on "Differentiation", "Benchmarks", "Industry Market", "Glossary"--evaluate technical barriers.
To Patients and Families
DiVo Gen²AI's CAR-T computing service is a business that works in coordination with medical institutions and CAR-T treatment teams, not a service directly for individual users and patients. CAR-T cell therapy involves serious immune reaction risks (e.g., CRS, ICANS) and must be implemented by professional teams in qualified medical institutions. We provide dry-lab computational steps for target screening and Binder design. Please contact us through your treating physician.
What is CAR-T
CAR-T stands for Chimeric Antigen Receptor T-cell Therapy. Simply put: T cells (the immune system's "warriors") are extracted from the patient, equipped with an artificially designed "navigator" (CAR), and reinfused to precisely identify and kill tumor cells.
The fundamental difference from traditional chemotherapy and targeted drugs is: CAR-T uses living cell drugs--after reinfusion, CAR-T cells proliferate, patrol, and continuously kill tumors in vivo, theoretically achieving long-term remission from a single treatment. Patients have survived cancer-free for over 10 years after a single CAR-T treatment.
But CAR-T is not a panacea. As of 2026, globally approved CAR-T therapies almost exclusively target hematological tumors (lymphoma, leukemia, myeloma). Solid tumors (lung, liver, gastric cancers accounting for 90% of cancers) remain the biggest challenge--the core bottleneck is target selection.
6 Steps of CAR-T Therapy
From blood collection to reinfusion, a complete CAR-T treatment workflow
Collection
Isolate T cells (a type of white blood cell, the immune system's "warriors") from patient blood
Engineering
Use viral vectors to introduce CAR genes into T cells, expressing chimeric antigen receptors on their surface--like equipping warriors with "navigators"
Expansion
Expand engineered CAR-T cells to hundreds of millions to billions in GMP facilities
Conditioning
Patient receives lymphodepletion chemotherapy to make "living space" for CAR-T cells
Infusion
Reinfuse CAR-T cells into the patient, where they begin seeking and attacking tumor cells expressing target antigens
Monitoring
Closely monitor adverse reactions including CRS (Cytokine Release Syndrome) and ICANS (Immune effector Cell-Associated Neurotoxicity Syndrome)
The entire process takes about 2-4 weeks (mainly cell engineering and expansion time). Step 2 "Engineering"--deciding who CAR targets--is the entry point for DiVo Gen²AI's computational service.
CAR Structure
The chimeric antigen receptor consists of 5 functional domains--DiVo primarily acts on the first
scFv / Binder
Antigen Binding DomainDiVoCAR's "eyes", recognizing target antigens on tumor surfaces. Traditionally uses scFv (single-chain variable fragment), new generation can use AI-designed protein Binders.
Hinge
HingeFlexible segment connecting the binding domain and transmembrane region, providing spatial flexibility.
Transmembrane
TransmembraneAnchors CAR to the T cell membrane.
Costimulatory
Costimulatory DomainProvides the second signal for T cell activation (CD28 or 4-1BB), determining CAR-T killing intensity and persistence.
CD3ζ
Signaling DomainProvides the first signal for T cell activation, triggering the killing mechanism.
CAR-T Development Milestones
From the first case in 2012 to the AI+Binder era in 2025
Emily Whitehead became the first child to receive CAR-T therapy, advanced ALL achieved complete remission
Cancer-free for 13 years
Kymriah (Novartis) received FDA approval, world's first CAR-T therapy launched
B-ALL indication
Yikaida (Fosun Kite) and Beinuoda (JW Therapeutics) received NMPA approval
China CAR-T year one
Carvykti (Legend Biotech/J&J) received FDA approval, Chinese-origin CAR-T goes global
BCMA multiple myeloma
Xie Qi/Cao Longxing Nature BME: computationally designed Binder replaces scFv, CAR-T efficacy significantly improved
AI+CAR-T milestone
Science trilogy: de novo pMHC binder can be integrated into CAR structure
CAR-T targeting intracellular neoantigens
5 Major Challenges in CAR-T
Each challenge corresponds to a computational entry point--see what DiVo can do
Solid Tumor Target Scarcity
Hematological tumors have mature targets like CD19/BCMA, but solid tumors lack tumor-specific surface antigens--targets are also expressed in normal tissue, causing "off-target toxicity".
Antigen Escape
Tumor cells evade CAR-T recognition by downregulating target antigen expression. Single-target CAR-T is prone to escape, requiring dual/multi-target strategies.
scFv Limitations
Traditional scFv derived from existing antibody libraries, with limited affinity and stability optimization space, and may trigger anti-drug antibody (ADA) responses.
Tumor Microenvironment Suppression
Immunosuppressive factors in the solid tumor microenvironment (TGF-β, IL-10, PD-L1) weaken CAR-T function.
Manufacturing Complexity
Autologous CAR-T requires "one person, one product", with a 2-4 week preparation period during which some patients experience disease progression.
DiVo Gen²AI's Role
In the CAR-T workflow, we handle the dry-lab computational steps of "navigator design"
In the CAR-T workflow, "deciding who CAR targets" and "designing the navigator" are the steps most dependent on computational prediction--impossible to experimentally validate each candidate target, algorithms must precisely filter from sequencing data.
- ▸Target Discovery--dual-channel parallel: VCF neoantigen screening + RNA-seq surface antigen mining
- ▸de novo Binder Design--RFdiffusion + ProteinMPNN, replacing traditional scFv
- ▸Structure Validation--Protenix atomic-level 3D structure, confirming interaction credibility
- ▸mRNA Sequence Optimization--GEMORNA optimizing CAR construct mRNA sequences
We do not produce CAR-T cells, do not provide clinical diagnostic opinions, do not do cell culture/expansion processes. We deliver target recommendation lists + Binder sequence design proposals that can directly enter wet-lab synthesis and preclinical validation.
5-Step End-to-End Pipeline
CT1-CT3 based on verified neoantigen pipeline, CT4-CT5 are new Binder design capabilities · Click steps to view underlying validation
HLA Typing + Variant Detection
✓ 已验证OptiType + Polysolver + GATK Mutect2 + VEP + ANNOVAR
4-digit HLA-I typing + somatic variant list
Neoantigen Screening + Surface Antigen Mining
✓ 已验证MHCflurry(14,883 alleles) + pVACseq + Salmon/DESeq2 differential expression
MHC high-affinity candidates + tumor-specific highly expressed membrane proteins
Full-Dimensional Immunogenicity Assessment
✓ 已验证5-dim scoring (affinity 30%+presentation 20%+processing 10%+immunogenicity 15%+structure 25%) + DeepTCR
Tier-graded candidate targets + TCR recognition feasibility
CAR Binder de novo Design
✓ 已验证PXDesign + RFdiffusion + ProteinMPNN
de novo protein Binder candidate sequences
Structure Validation + Scoring
✓ 已验证Protenix(pLDDT/ipTM) + Protenix-Dock + PXMeter
Atomic-level complex structure + docking score + design quality assessment
Three Key Differentiators
Capabilities that traditional CAR-T pipelines lack
Dual-Channel Target Discovery
Traditional CAR-T target discovery relies on mutation annotation or experience. DiVo runs two channels simultaneously--VCF neoantigen screening (MHC affinity) + RNA-seq differential expression (tumor-specific membrane proteins), cross-validating to produce candidate targets.
de novo Binder Design
Traditional CAR-T uses scFv antibody fragments as antigen binding domains, limited by existing antibody libraries. DiVo uses RFdiffusion + ProteinMPNN to design protein Binders from scratch, breaking through scFv affinity and stability bottlenecks--Xie Qi/Cao Longxing 2024 Nature BME validated this approach.
pMHC 3D Structure Validation
Traditional pipelines stop at IC50 values. DiVo builds atomic-level pMHC 3D structures for each candidate target (pLDDT=95.4, ipTM=0.977), validating antigen-MHC interaction credibility at the structural level, reducing false positives.
Verifiable Engineering Foundation
| Metric | Value | Note |
|---|---|---|
| MHC-I allele coverage | 65+ | IC50 min 10.1 nM |
| MHC-II allele coverage | 25 | 13 DR + 5 DP + 7 DQ |
| pMHC structure pLDDT | 95.4 | Approaching X-ray crystallography precision |
| pMHC ipTM | 0.977 | Complex overall confidence |
| RNA-seq index | 253,181 | GENCODE v46 transcripts |
| TCR repertoire | TRUST4 v1.1.9 | Clonotype diversity analysis |
| Binder design pipeline | RFdiffusion + MPNN | From backbone to sequence end-to-end |
| Structure validation | Protenix + PXMeter | 400+ complexes validated at scale |
Verified Technical Capabilities
Each underlying pipeline has independent validation data + sample reports
Tumor Neoantigen mRNA Vaccine 8-Step Pipeline
✓ 已验证· Has sample reportUnderlying pipeline for CT1-CT3. 8-step end-to-end verified, TESLA AUROC=0.698, 45 candidates 17 Tier-1.
pMHC 3D Structure Prediction + Docking Validation
✓ 已验证· Has sample reportCore capability of CT5. Protenix + Protenix-Dock, pLDDT=95.4, ipTM=0.977, 400+ complexes validated.
HLA Typing (Dual-Tool Cross-Validation)
✓ 已验证· Has sample reportCore capability of CT1. OptiType + Polysolver dual-tool, 4-digit resolution HLA-I typing.
mRNA Sequence Optimization (RNALens)
✓ 已验证· Has sample reportCAR construct mRNA sequence optimization. Spearman=0.92, CAI 0.7->0.95. Self-tuned, not API call.
Genome Interpretation Report
✓ 已验证· Has sample reportWGS end-to-end interpretation, PRS + TMB + pharmacogenomics, bilingual. Can be used for CAR-T pre-treatment genomic risk stratification.
Honest Boundaries
What we can and cannot do, clearly stated
What We Can Do
What We Don't Do
Industry Market Reference
Compiled from public information · Not DiVo pricing · For reference only
Boundary Note: The following are market public reference prices for complete courses of CAR-T cell therapy, covering cell collection, CAR engineering, expansion, reinfusion, hospitalization, and adverse reaction management. DiVo Gen²AI handles the dry-lab computational steps (i.e., the 5-step pipeline on this page), and is not the producer or treatment provider of CAR-T cells.
Solid tumor CAR-T is the global R&D main battlefield--whoever first solves target selection and off-target toxicity will capture the next decade's growth engine. The global CAR-T market is projected to exceed $50 billion by 2030.
Glossary
10 most common terms in CAR-T
| Abbr. | Full Name | Translation | Explanation |
|---|---|---|---|
| CAR | Chimeric Antigen Receptor | Chimeric Antigen Receptor | Artificially designed receptor enabling T cells to recognize specific target antigens |
| scFv | Single-Chain Variable Fragment | Single-Chain Variable Fragment | Traditional CAR antigen binding domain, derived from natural antibodies |
| CRS | Cytokine Release Syndrome | Cytokine Release Syndrome | CAR-T activation releases large amounts of cytokines, causing fever, hypotension, potentially life-threatening |
| ICANS | Immune effector Cell-Associated Neurotoxicity Syndrome | Immune effector Cell-Associated Neurotoxicity Syndrome | Neurotoxicity reaction from CAR-T therapy, presenting as confusion, seizures, etc. |
| Neoantigen | Neoantigen | Neoantigen | Antigen peptides produced by tumor-specific mutations, not expressed in normal tissue |
| MHC | Major Histocompatibility Complex | Major Histocompatibility Complex | The "display board" on cell surfaces, presenting intracellular protein fragments to T cells |
| pMHC | peptide-MHC complex | peptide-MHC Complex | Antigen peptide complex presented by MHC molecules, the target of T cell recognition |
| Binder | De novo Protein Binder | De novo Protein Binder | AI-designed proteins from scratch that can bind targets with high affinity, replacing traditional scFv |
| TMB | Tumor Mutational Burden | Tumor Mutational Burden | Number of mutations in the tumor genome, high TMB usually means more neoantigen candidates |
| ipTM | interface predicted TM-score | interface predicted TM-score | Protein complex interface interaction confidence, >0.75 is high confidence |
CAPACITY TRACE
能力回溯
这项服务由哪些能力支撑——从硅片到你的场景
硅片(L1) → 模型(L2) → Agent(L3) → 管线(L4) → 你的场景