From 766ab52f2458b79124bcc4866a0bda17aec28de6 Mon Sep 17 00:00:00 2001 From: Terence Lim Date: Wed, 1 Jul 2026 20:49:27 +0800 Subject: [PATCH] update docs --- source/rest-api/embeddings.rst | 2 ++ 1 file changed, 2 insertions(+) diff --git a/source/rest-api/embeddings.rst b/source/rest-api/embeddings.rst index 8358b5c..af679e1 100644 --- a/source/rest-api/embeddings.rst +++ b/source/rest-api/embeddings.rst @@ -24,6 +24,8 @@ Currently, we support the following models: - **ProteinMPNN**: A deep learning model for sequence design conditioned on a fixed protein backbone. It predicts amino acid sequences likely to fold into a given 3D structure by modeling residue–residue interactions through message passing on the protein’s spatial graph. Used for structure-based protein design, stability optimization, and inverse folding tasks. `GitHub link `__, `Reference `__. Licensed under `MIT `__. +- **SolubleMPNN**: A variant of ProteinMPNN trained with weights tuned for soluble protein targets. Uses the same inverse-folding pipeline as ProteinMPNN; `soluble_model` is implied by the route, and `model_name` is restricted to the soluble-compatible weights (`v_48_010`, `v_48_020`). `GitHub link `__, `Reference `__. Licensed under `MIT `__. + - **ESM-IF1**: An inverse-folding model pretrained on PDB and ~12 million AlphaFold-predicted structures. Given a backbone structure, it scores and generates amino acid sequences predicted to fold into it, making it another option for structure-based design and inverse folding alongside ProteinMPNN. `GitHub link `__, `Reference `__. Licensed under `MIT `__. - **AbLang2**: An antibody-specific language model that provides residue-level and sequence-level representations for both heavy and light chains. `GitHub link `__, `Reference `__. Licensed under `BSD-3 `__.