Medvolt

Generative models, knowledge graphs and language models drive the core modules: Topaaz proposes new chemical entities, Rubie ranks repurposing candidates and targets, and EDGE turns literature into a structured knowledge graph that the models reason over. Physics based methods, including free energy perturbation in Oopal, molecular dynamics and docking, score and refine what the models propose, and expert teams review outputs before delivery.

Part of the value sits in those simulations and in curation services that stand apart from the learned models.

Models generate and rank candidates, physics based simulations test them in silico, and multidisciplinary teams review outputs before they reach a client, so a person stands between the platform and the delivered result.

How that review works, and what makes a candidate fail it, is not described.

The platform combines generative models, natural language processing, transfer learning, reinforcement learning and large language models with an ontology based knowledge graph, and pairs them with free energy perturbation, molecular dynamics and deep docking. Models are described as interpretable and trained on curated domain data.

No model, architecture or training dataset is named.

Medvolt's own models and knowledge graph, built on curated proprietary data, are combined with large language models, and the company lists AWS Startups, NVIDIA Inception and Microsoft for Startups among its partner programs.

Which language models and cloud providers process client data is not named.

Medvolt says its platform cuts preclinical discovery time threefold, reduces costs fifteenfold and lowers failure risk by 25 percent, and that its modules are refined through laboratory testing. Ten case studies describe projects in small molecules, repurposing, knowledge curation and enzyme engineering.

The figures carry no source, method or named client.

MedGraph works on molecular, literature and multi omics research data, and its knowledge curation services annotate genetic variants and gene, drug and disease relationships from published sources.

No privacy policy or statement on how client data is handled appears on the site.

The platform handles research data for discovery projects rather than patient records, and Medvolt sells to pharmaceutical and biotechnology research teams.

No statement on patient data or business associate status is published.

The platform is described as secure and enterprise class, with cloud and on premise options built to be infrastructure agnostic.

No certification, security page or outside assessment is named.

MedGraph is research software for early discovery, used by Medvolt's scientists and by client research teams, and makes no device claim. Its outputs are candidate molecules and target hypotheses that enter laboratory work rather than patient care.

Any candidate that reaches the clinic would follow the drug pathway under its developer.

Models are described as interpretable and trained on carefully curated data, and the user interface shows outcomes across the discovery pipeline.

No evaluation of model behavior, by target class or otherwise, is published.

Expert teams review model outputs before delivery, and candidates are checked by physics based simulation before any laboratory work, which gives a client a reviewed result rather than raw model output.

No accuracy figure, warranty or remedy covers a wrong prediction.

MedGraph works with molecular structures, literature and omics data inside its own workflows, so no clinical record system is involved, and clients reach it through the platform interface.

No integration with laboratory notebooks or research data systems is described.

MedGraph runs as a cloud native platform where clients build workflows, launch high performance computing experiments and collaborate, and it can also be installed on premises in a client's own environment.

Where the cloud version runs and where client data rests is not stated.

Medvolt sells MedGraph as modular software and as customized projects delivered by its scientists, positioned as affordable for organizations of different sizes, with demo and contact routes on every page and downloadable service brochures.

No list price, rate or contract term appears on the site.

The platform covers small molecules, peptides, antibodies, oligonucleotides and enzymes, with oncology, autoimmune, metabolic, neurodegenerative and rare diseases named as therapeutic areas. Case studies describe kinase and DNA repair inhibitors, bacterial target prioritization, drug repurposing and enzyme engineering for biocatalysis.

None of the case studies names a client or reports results by therapeutic area.