Ainnocence launches protein, antibody and peptide AI models as APIs

Ainnocence, Inc. has moved three proprietary drug-discovery AI models into production as secure, on-demand APIs that partners can plug into existing pipelines. The launch turns benchmarked protein, antibody and peptide predictions into live services aimed at speeding discovery and keeping the underlying models protected. Why it matters: - Ainnocence is making its core AI models usable inside pharmaceutical and biotechnology workflows without a custom research integration. - The API launch is designed to shorten the path from sequence input to prediction, which can speed candidate triage and downstream discovery decisions. - The model-as-a-service setup also keeps Ainnocence’s underlying models protected while partners use the outputs in their own systems. What happened: - Ainnocence, Inc. announced the production launch of three proprietary models as secure, on-demand APIs on June 12, 2026. - The live models are AINN-P1, AINN-Basemodel and AINN-Peptide. - The company positioned the release as a direct production path for pharma and biotech partners to integrate predictions into discovery pipelines. - Dr. Lurong Pan, founder and CEO of Ainnocence, said the API approach lets partners access predictive power from their own pipelines without exposing the underlying science. The details: - AINN-P1 is a protein foundation model trained on UniRef sequences and served in three sizes: 167M, 500M and 1B parameters. - AINN-P1 returns high-dimensional sequence embeddings for downstream property prediction and research-stage sequence scoring and generation. - AINN-Basemodel predicts antibody-antigen binding affinity with a graph-neural-network ensemble and scores results on a 0-to-1 scale. - A dedicated batch endpoint for AINN-Basemodel can rank roughly 1,000 antibody-antigen pairs in a single request. - AINN-Peptide classifies peptides and profiles them across 13 therapeutic activities, including antimicrobial, anticancer, antiviral and immunomodulant. - All three models use one consistent interface that accepts a protein or peptide sequence, or antibody chains, and returns structured predictions. - The APIs support single-sequence predictions in well under a second, with many responses taking tens of milliseconds. - The platform also supports high-throughput batch screening at the multi-million-variant scale. - Capacity scales elastically with demand, so partners can move from one query to large screening campaigns without changing integration. - Each partner gets dedicated API credentials and usage tiers matched to expected demand. - The APIs are built to fit existing computational pipelines, electronic lab notebooks and cloud workflows. - Ainnocence said adding its predictions to automated screening or ranking workflows takes only a few lines of code. Between the lines: - The release turns Ainnocence’s earlier benchmark results into a commercial delivery channel. - Ainnocence cited prior results for AINN-P1 on the ProteinGym leaderboard and said its peptide platform screened more than 3.2 million variants at greater than 90% activity-classification accuracy. - The company is emphasizing both speed and control: fast access for partners, but no exposure of the models themselves. - That mix suggests Ainnocence is aiming to make adoption easier for discovery teams that already have established software stacks. What’s next: - Ainnocence is inviting pharmaceutical and biotechnology partners to integrate the APIs into ongoing discovery programs. - The company is expected to use the launch to expand use of its protein, antibody and peptide models across more workflows. - The API model also creates a clearer path for scaling from individual queries to larger screening campaigns. The bottom line: - Ainnocence has converted three of its drug-discovery AI models from research assets into production services that partners can call on demand.