From toxicology model to product: building DioScor’s usable front end

Consone AI had already assembled curated toxicity data from human and animal studies, developed the deep-learning capability behind organism-level predictions, and defined where DioScor should sit as a product. The model and proposition were not the missing pieces.

What remained was the software surface through which somebody could submit a compound, understand the result, inspect confidence and supporting evidence, and move through a coherent reporting journey without losing the scientific nuance underneath it.

A scientific AI interface cannot stop at presenting a prediction. DioScor needed to communicate cross-species and organ-level findings, confidence, structural attribution, biomarkers, mechanisms and known evidence without collapsing them into a false single answer.

The product also needed to distinguish curated source data from model output and keep the limits of prediction visible. DioScor supports interpretation; it does not replace regulatory testing or guarantee that a compound is safe.

Veda translated Consone AI’s product direction and subject-matter expertise into the front-end architecture: the information model, interaction flow, visual hierarchy and responsive implementation needed to make the capability usable.

The resulting journey moves from molecular input and exposure route through prediction summary, cross-species comparison, structural and mechanistic interpretation, weight of evidence and exportable reporting.

DioScor’s interface is deliberately layered. A user can begin with the prediction summary and confidence, then move into organ-level risk, species comparison, molecular regions, biomarker pathways, mechanisms and similar compounds as more scrutiny is required.

That structure makes technical depth available without placing every result at the same visual level. It also treats explainability as a core interaction requirement rather than a disclaimer attached after the prediction.

The weight-of-evidence view places a result alongside known compounds, molecular similarities, relevant species, endpoints and applicability information. Instead of asking a user to accept an isolated score, the product gives them a route into the material that supports interpretation.

That is especially important in scientific software: trust comes from showing how a result can be examined, where its confidence comes from and which questions remain outside the model’s scope.

Consone AI supplied the science, predictive model and product direction. Veda’s role was to turn those defined requirements into a structured, visual and reportable front-end experience, then deliver the responsive interface needed for demonstration, review and launch.

Front-end delivery is complete and the product is awaiting go-live. Until live evidence exists, this case study does not claim adoption, commercial performance or scientific outcomes. The proof is the product translation and implementation itself.