Contact

The bottleneck has shifted

The challenge is choosing which edits belong together.

Multiplex editing has made higher-order cell design technically feasible. KiraLOGIC combines human disease evidence, a pretrained phenotype model, and KiraGen's experimental data to prioritize complete architectures for testing.

~80 Biology-selected genes
4–8 edits Current operating range
>30 billion Possible combinations within that range

As multiplex capacity expands, architecture selection becomes more important, not less.

Approximately 80 biology-selected genes expand into a current four-to-eight-edit higher-order search space. KiraLOGIC ranks a focused set of complete architectures for experimental testing, while the design space continues beyond the current operating range.

Complete-product learning

The product is the combination.

Broad screens expand the parts list. KiraLOGIC asks which selected edits work together in the same finished CAR-T product.

Each prioritized architecture is generated as a separate multiplex build and linked from intended architecture through realized product state and measured phenotype under tumor pressure.

Three separately generated multiplex CAR-T architectures are each linked from intended architecture through realized product state and measured phenotype under tumor pressure.

Build A

Intended architecture
Realized product state
Measured phenotype

Build B

Intended architecture
Realized product state
Measured phenotype

Build C

Intended architecture
Realized product state
Measured phenotype

The learning system

Every build sharpens the next decision.

Deeply characterized cell-therapy experiments are limited in number. Phenomena provides a pretrained large phenotype model that captures relationships among cellular states, perturbations, and measured phenotypes.

KiraGen combines that model with exact-product records and active experimental design to prioritize builds by predicted performance, uncertainty, and diversity. Matched architecture, product-state, model-context, and phenotype records inform the next ranking.

The model prioritizes. The data decide.

A pretrained Phenomena large phenotype model and KiraGen exact-product data feed KiraLOGIC, which prioritizes the next informative higher-order architectures. Selected builds fan into defined suppression for controlled comparison, patient-derived spheroids for robustness across patient tumors, patient-derived tumoroids with retained human tumor-microenvironment context, and orthotopic PDX models for intracranial control over time. Matched architecture, product state, model context, and phenotype records return to KiraGen's dataset to refine subsequent rankings.

Model + active design

Pretrained large phenotype model

Phenomena · Reflector Bio

Cell state+Perturbation+Phenotype
Pretrained across millions of heterogeneous perturbation experiments

KiraGen exact-product data

Architecture · product state · model context · phenotype
Adaptive selection system

KiraLOGIC

Large phenotype model + product-grounded active design

Performance · Uncertainty · Diversity

Next informative builds

Selected to be promising, informative, and different

Measure across complementary biology

Scalable comparative learningHigh-fidelity translational learning

Defined suppression

Controlled, scalable comparison

Patient-derived spheroids

Robustness across patient tumors

Patient-derived tumoroids

Retained human TME context

Orthotopic PDX

Intracranial control over time

Defined suppression

Controlled, scalable comparison

Patient-derived spheroids

Robustness across patient tumors

Patient-derived tumoroids

Retained human TME context

Orthotopic PDX

Intracranial control over time

Illustrative biological contexts

Prospective higher-order closed-loop performance remains under validation.