ATGO is a deep learning-based algorithm for high accuracy protein Gene Ontology (GO) prediction.
Starting from a query sequence, it first extracts three layers of feature embeddings
from a pre-trained protein language model (ESM-1b).
Next, a fully connected neural network is used to fuse the feature embeddings, which
are then fed into a supervised triplet network for GO function prediction.
Large-scale benchmark tests demonstrated significant advantage of ATGO
on protein function annotations due to the integration of discriminative feature embeddings
from attention transformer models.
(view an example of ATGO prediction)
ATGO On-line Server
ATGO Download
References:
-
Yi-Heng Zhu, Chengxin Zhang, Dong-Jun Yu, Yang Zhang.
Integrating unsupervised language model with triplet neural networks for protein gene ontology prediction.
PLOS Computational Biology, 2022, 18 (12): e1010793.
umich.edu
| (734) 647-1549 | 100 Washtenaw Avenue, Ann Arbor, MI 48109-2218