VIDRAFT's PharmaOS Lands Government-Pharma Program to Drive AI-Discovered Small-Molecule Cancer Drug Candidates to Preclinical
TL;DR: VIDRAFT, a Korean AI-foundry deep-tech company, has been selected for the South Korean government's "Challenge Bio" open-innovation program to discover and validate small-molecule anticancer drug candidates using its PharmaOS platform — running from hit identification through preclinical linkage, with IP acquisition as an explicit goal. The platform has previously ranked #1 across 16 categories on the global Polaris drug-prediction benchmark. Developers and researchers can already interact with VIDRAFT's open drug-discovery ecosystem on Hugging Face.
What it is
VIDRAFT (비드래프트) is a Korean "AI Foundry" startup — its business model centers on designing, producing, and auditing AI models for industrial deployment, rather than being a pure software SaaS play. The core drug-discovery product is PharmaOS, a computational platform for:
- Molecular structure exploration — searching chemical space for candidate compounds
- Target-protein binding analysis — estimating likelihood of binding to a specified protein target
- Drug-likeness and toxicity risk assessment — filtering candidates before wet-lab work begins
- Explainable candidate prioritization — the platform emphasizes why a compound advances, not just that it scored highly
The broader research stack also includes:
- QuantumOS — a quantum-classical hybrid computation software layer being researched for molecular-level analysis
- CELLOS — a cell and organ simulation system under active development, designed to model cellular metabolism and inter-organ interactions, providing hypotheses for candidate comparison and downstream experiment design
Together, these form an automated, iterative discovery-and-validation loop: AI proposes candidates → quantum/classical hybrid layer refines molecular analysis → cell/organ simulation flags biological risks → findings feed back into molecular design.
How it works
PharmaOS operates as a multi-stage computational funnel rather than a single-score ranker:
- Generation — The system explores molecular structure space to produce a large pool of candidate compounds against a specified oncology target.
- Binding estimation — Computational docking and affinity models assess each candidate's likelihood of interacting with the target protein.
- ADMET profiling — Drug-likeness, absorption, distribution, metabolism, excretion, and toxicity risks are estimated in silico.
- Rationale generation — Rather than outputting only a ranked list, PharmaOS produces documented reasoning for why specific candidates merit wet-lab follow-up — an important practical feature when handing off to pharma partners.
- Iterative refinement — Results from each computational stage feed back into the design loop, with QuantumOS research exploring additional molecular-level verification and CELLOS modeling biological responses at the cellular and organ level.
The pipeline is explicitly designed to connect open, standardized computational evaluation to real industrial research demands — the goal being compound IP, not just a proof-of-concept publication.
Benchmarks & results
- Polaris benchmark: PharmaOS has publicly claimed #1 rankings across 16 categories on Polaris, the global drug-prediction benchmark platform, making it one of the more independently verifiable performance claims in the Korean AI-pharma space.
- Open Discovery Challenge (ODC): VIDRAFT runs a public, open drug-discovery competition on Hugging Face in both Korean and English. As of September 6, 2026, the platform has scored a cumulative 13,067 submissions across four disease-area seasons: malaria, tuberculosis, Chagas disease, and non-opioid analgesia. Submissions are evaluated against published, season-specific computational criteria — anyone can submit and compare results on an equal basis.
- Program selection: VIDRAFT was chosen through a competitive multi-stage evaluation (document screening → meetup → presentation review) from among entrants competing for the research demands posted by 17 pharmaceutical companies, including Dong-A ST, Yuhan Corporation, Hanmi Pharmaceutical, SK Chemicals, and HK Inno.N.
How to try it
VIDRAFT's Open Discovery Challenge (ODC) is publicly accessible on Hugging Face in both English and Korean. Developers and computational chemists can:
- Browse open challenge seasons (malaria, TB, Chagas disease, non-opioid pain)
- Submit AI-designed molecules and receive computational scoring under publicly posted evaluation criteria
- Compare results against all other submissions on equal terms
The platform is available at VIDRAFT's Hugging Face organization page. No private API access or invitation is required for ODC participation.
Note: PharmaOS itself, the QuantumOS layer, and CELLOS are not described as publicly self-hostable at the time of this writing. The ODC on Hugging Face is the primary developer-facing entry point.
FAQ
Q: What oncology targets is VIDRAFT working on under this program?
A: Specific targets are determined by agreement between VIDRAFT and the participating pharmaceutical companies under the finalized project plan. The source article does not disclose them, and they will likely be governed by the IP agreements being established.
Q: Is the ODC scoring methodology published openly?
A: Yes — the evaluation criteria are published per season on the ODC platform, and all participants are scored against the same public criteria. This is explicitly part of VIDRAFT's design: open problems, open scoring, comparable results.
Q: What is the "Challenge Bio" program, and who funds it?
A: "모두의 챌린지 바이오" (Challenge Bio for All) is a South Korean Ministry of SMEs and Startups open-innovation program that matches startup technology with pharma company research demands. Funding combines government grants with support from the participating pharmaceutical companies. The Korea Innovative Medicine Consortium (KIMCo) cooperates on the program.
Q: Where does QuantumOS fit — is this real quantum hardware?
A: QuantumOS is described as a quantum-classical hybrid software layer under research for molecular-level analysis. The source article describes it as a research direction complementing PharmaOS outputs — no hardware deployment details are disclosed.
Originally reported by 바이오타임즈 (2026-09-07) — source article.
Top comments (0)