Expanding Therapeutic Target Space from Genes to Disease-Relevant Molecular Forms with Agent3Diso

Therapeutic targets are commonly defined at the gene or canonical protein level, yet therapeutic effects are ultimately exerted through specific molecular states operating in specific biological contexts. Alternative splicing generates diverse RNA and translated protein forms whose functional consequences, disease relevance, and therapeutic vulnerabilities can differ substantially within the same gene. Resolving these disease-relevant molecular forms creates opportunities for selective intervention across complementary RNA- and protein-directed modalities, but requires reasoning across tightly coupled molecular layers, heterogeneous evidence, explicit biological constraints, and modality-specific therapeutic requirements. Here, we present Agent3Diso, a hybrid deterministic--agent multimodal framework for molecular-form-resolved therapeutic discovery. Agent3Diso establishes cross-layer molecular coherence from transcript identity and translation to structure, function, and disease context; enforces explicit molecular and therapeutic constraints through deterministic computation; and applies specialized agentic reasoning to evidence synthesis, mechanistic interpretation, conflict resolution, hypothesis critique, and experimental design. The framework evaluates therapeutic opportunities across multiple intervention modalities and translates prioritized disease-relevant molecular states into traceable hypotheses linking molecular identity, therapeutic rationale, modality, concrete intervention, uncertainty, and experimental validation. By integrating molecular resolution with constraint-aware and modality-conditioned agentic reasoning, Agent3Diso provides a framework for extending therapeutic discovery beyond gene-level targets toward disease-relevant molecular forms.