At the 2026 World Artificial Intelligence Conference (WAIC), the AI biotechnology firm Molecular Heart unveiled new findings from its AI drug design platform, MMDesign, focusing on the de novo design of cyclic peptides. The platform's designed candidate cyclic peptides for protein-protein interaction (PPI) targets, such as p53-MDM2 and PD-L1, have been validated through wet-lab experiments.
PPI targets are crucial for research into diseases like cancer and immune disorders. However, their large and intricate binding interfaces make traditional small-molecule drug development challenging. Cyclic peptides have emerged as a promising avenue in innovative drug R&D due to their smaller size and potential to engage complex targets. Yet, their sequence, cyclization method, and 3D conformation are interdependent, and conventional development often relies on costly, large-scale peptide library screening.
The data released by Molecular Heart shows that for the p53-MDM2 target, 16 out of the 21 candidate cyclic peptides designed and tested in the first round by MMDesign showed experimental hits, representing a success rate exceeding 70%. Among these, the optimal candidate demonstrated a binding affinity (KD) of 13.7 micromolar, outperforming the positive control used in concurrent experiments. For the PD-L1 target, 6 out of 19 AI-designed candidate cyclic peptides in the first round were confirmed to bind the target via surface plasmon resonance (SPR) assays, with the best KD nearing 40 micromolar. Compared to globally reported results for AI-driven de novo peptide design, these first-round hit rates and binding activity data for two classic PPI targets position the work at a leading international level.
Key Development
Perhaps more significant is the shift in research methodology this outcome represents. It demonstrates that AI can substantially narrow the candidate space before wet-lab experiments begin, allowing R&D teams to prioritize limited experimental resources on molecules with higher binding potential. For molecular modalities like cyclic peptides and others with vast candidate spaces and high screening costs, this "generate-screen-validate" pathway holds promise for reducing time and resource investment in early discovery stages. It also offers a new technical route for drug discovery targeting complex mechanisms like PPIs.
Just one month prior, Molecular Heart announced results for AI-driven de novo design of nanobodies. Across over a dozen real therapeutic targets, MMDesign consistently generated molecules with nanomolar to picomolar affinity while limiting candidate designs to fewer than 50 per single target. Nanobodies and cyclic peptides differ significantly in molecular size, structural features, and design constraints. Achieving validated de novo design in both modalities within a short timeframe is a key metric for assessing the generalization capability of an AI biological design platform and constitutes a relatively scarce technical competency for companies in this field.
Founder's Perspective
Molecular Heart founder Xu Jinbo stated that the key to AI-driven biologic drug development lies not in a single generation or prediction result, but in establishing a "generate-compute-experiment-iterate" closed loop. This process continuously calibrates the model's understanding of structure, interactions, and function using real experimental data. The field of AI for Science (AI4S) is evolving from competition in single-point models and individual tasks toward a competition in systematic scientific discovery capabilities.
Platform Accessibility
The relevant capabilities of MMDesign have been integrated into Molecular Heart's self-developed AI macromolecular design operating system, MoleculdOS. The company plans to gradually open these capabilities for collaborative partnerships with pharmaceutical firms, universities, research institutes, and innovative teams.
Comments