Insilico Medicine

GPT-6

2026-09-11 04:23:10

GPT-6 Astra tops DDD benchmark in antibody developability test

A post amplified by OpenAI President Greg Brockman has pushed GPT-6 Astra into the center of discussion across AI and biopharma. Scientist Andrew Aiginin said on X that GPT-6 Astra outperformed all frontier models in an independent benchmark for antibody developability prediction, making it the top AI system on that task. According to the claim, Astra scored 37.98 on the antibody developability module of the DDD Benchmark, which combines six different types of antibody experimental data, and also built an interactive visualization page for complex antibody mechanisms in about one hour. The result was discussed alongside a 21-page arXiv paper from Insilico Medicine, titled "Training Chemical Plausibility-Aware Large Language Models for Single-Step Retrosynthesis." The paper describes the broader evaluation philosophy behind the DDD Benchmark and details methods used in small-molecule chemistry, including the URSA-expert-2026 benchmark, the ChemCensor scoring framework, a dataset of roughly 45.6 million validated reactions, and GRPO-based reinforcement learning with a novelty reward. In the article’s framing, GPT-6 Astra’s performance suggests that general-purpose large models are moving beyond code and literature support into harder preclinical prediction work in drug discovery.

1520

Insilico Medi

2026-09-09 10:12:57

Insilico paper says AI-designed fibrosis drug cut biological age readings within four weeks

Insilico Medicine said in a paper published Sept. 7 in Nature Biotechnology that its drug candidate rentosertib reduced biological age readings across six independent proteomic aging clocks in blood samples from 42 patients with idiopathic pulmonary fibrosis. The strongest signal appeared in the 30 mg twice-daily group, where biological age was reported to move back by about three to four years at week four, with one clock showing a six-year reversal. The study framed the analysis as the first time multiple proteomic aging clocks had been compared within a clinical drug intervention trial. Founder Alex Zhavoronkov called it the most important paper of his career so far and said the work was first presented on Sept. 8 at a Nature conference at Sorbonne University in Paris. The paper also sets clear limits on what can be concluded. The 42-patient dataset was an exploratory subset of a 71-patient phase 2a trial, and the authors said the study cannot yet separate slower aging from the possibility that the blood-based changes simply reflect improved lung disease. The company’s earlier Nature Medicine paper highlighted a different result: the clearest lung-function gain came from the 60 mg once-daily group, not the 30 mg twice-daily group that showed the largest aging-clock change.

1270

Insilico Medi

2026-07-17 08:28:09

Insilico Medicine CEO says AI plus China’s R&D structure can cut drug discovery to 9-12 months

Insilico Medicine founder and CEO Alex Zhavoronkov said at the 2026 World Artificial Intelligence Conference that artificial intelligence is reshaping early-stage drug discovery. He said the traditional process from target identification to a candidate compound usually takes four to five years. China’s domestic R&D structure and talent base alone can shorten that timeline by about two years versus the global average, and adding AI can reduce it further to nine to 12 months. Zhavoronkov also pointed to the company’s earlier work, saying a 2019 Nature paper from his team showed that reinforcement learning and AI could generate a new molecule and validate it in the lab within 46 days. He added that the spread of generative AI, along with open models such as DeepSeek, has changed research workflows and made AI more accessible for public health and drug development. He disclosed that Insilico already has drug programs in Phase III clinical trials in China and said the company has built a robot-driven drug discovery facility in Jinqiao, Pudong, Shanghai. The facility went from concept to completion in 18 months, and the company is now offering that capability to other pharmaceutical firms across more than 1,200 tasks involved in new drug development.

2420