How does a 167M-parameter protein foundation model outperform a 650M general-purpose model on the tasks that matter?
167M
Parameters
0.625
ProteinGym Stability
+0.15
AUC vs ESM2 650M
3.9×
Fewer Parameters
Ainnocence Inc., an AI-driven drug discovery company, today announced new results for AINN-P1, a purpose-built protein foundation model designed to deliver efficient and transferable predictions for protein engineering and biologics discovery.
AINN-P1 takes a sequence-first approach, generating protein representations directly from amino acid sequences without requiring multiple sequence alignments, structure prediction, or external functional annotations. Conventional pipelines rely on homolog search, which is slow and requires large databases, while AlphaFold-class structure prediction still costs minutes per sequence.
The encoder uses a multiplicative LSTM rather than a Transformer. Where all-to-all attention requires n² connections and a key-value cache that grows with sequence length, recurrent state passing requires n steps at a memory footprint that does not grow at all — linear O(n) time complexity against the Transformer's quadratic O(n²). The model is trained autoregressively, predicting the next amino acid, rather than with the masked language modeling objective used by the ESM family.
Transformer all-to-all attention requires n² connections; mLSTM recurrent state passing requires n steps, with a memory footprint that does not grow with sequence length.
Leading Stability Prediction on ProteinGym
ProteinGym consolidates deep mutational scanning assays across four task categories. AINN-P1 leads on stability — a gating property for biologics developability.
Model
Params
Stability
Activity
Binding
Expression
Avg ρ
AINN-P1
167M
0.625
—
—
—
0.441
ProSST (structure-aware)
—
~0.590
gap 0.03–0.06
gap 0.02
gap 0.03–0.06
—
xTrimoPGLM
100B
~0.450
—
—
—
—
AINN-P1 stability score is +6% vs ProSST and +39% vs the 100B xTrimoPGLM. Gaps for activity/binding/expression are 0.02–0.06 in AINN-P1's favor — sequence alone encodes more of the relevant constraint than commonly assumed.
Testing Generalization on New Antibody Programs
Antibody discovery data are organized into programs sharing large stretches of framework sequence. Under a random split, close relatives land on both sides — a model can score well just by learning program identity. Under leave-program-out (LPO), only transferable biophysics can help.
Ainnocence evaluated VHH single-domain antibody expression under both protocols on proprietary data from real discovery programs.
Encoder
Params
Random Split AUC
Leave-Program-Out AUC
AUC Drop (leakage diagnostic)
AINN-P1
167M
≈ same
0.810
minimal
ESM2 (finetuned)
650M
≈ same
0.804
small
ESM2-base (no finetuning)
650M
0.882
0.660
−0.222
Key Result
AINN-P1 beat the 650M ESM2 baseline by +0.15 AUC on unseen programs (0.810 vs 0.660) with 3.9× fewer parameters, and matched the finetuned ESM2 within 0.006 AUC — without any task-specific finetuning.
Ablation across 7 classifiers × 3 encoders (21 configs): ESM2-base was weakest at every classifier position — the effect is a property of the embeddings, not the head fitted on top.
"The most important result for us is not simply that AINN-P1 performs well on a benchmark. It is that the model maintains strong performance when evaluated on antibody programs it has not seen before. That distinction is critical for drug discovery, where the real test is whether a model can support decisions on the next program, not just reproduce patterns from the last one."
Founded in 2021 and headquartered in California, Ainnocence is a next-generation biotechnology company transforming drug discovery and synthetic biology through AI-based, sequence-first engineering. The company's self-evolving platform evaluates up to 10 billion molecules spanning proteins, antibodies, small molecules, nucleic acids, and chemical formulations within hours to weeks, enabling rapid, multi-objective design across therapeutic, biological, and chemical systems. By reducing R&D timelines and costs while increasing success rates, Ainnocence empowers industry and academic partners to pursue complex biological innovation with greater precision and control.