The Gene Ontology (GO) is a formal, axiomatic theory with over 100,000 axioms that describe the molecular functions, biological processes and cellular locations of proteins in three subontologies. Predicting the functions of proteins using the GO requires both learning and reasoning capabilities in order to maintain consistency and exploit the background knowledge in the GO. Many methods have been developed to automatically predict protein functions, but effectively exploiting all the axioms in the GO for knowledge-enhanced learning has remained a challenge. We have developed DeepGO-SE, a method that predicts GO functions from protein sequences using a pretrained large language model. DeepGO-SE generates multiple approximate models of GO, and a neural network predicts the truth values of statements about protein functions in these approximate models. We aggregate the truth values over multiple models so that DeepGO-SE approximates semantic entailment when predicting protein functions. We show, using several benchmarks, that the approach effectively exploits background knowledge in the GO and improves protein function prediction compared to state-of-the-art methods.

基因本体 (GO) 是一种正式的公理理论，拥有超过 100,000 条公理，描述了三个子本体中蛋白质的分子功能、生物过程和细胞位置。使用 GO 预测蛋白质的功能需要学习和推理能力，以保持一致性并利用 GO 中的背景知识。人们已经开发出许多方法来自动预测蛋白质功能，但有效利用 GO 中的所有公理进行知识增强学习仍然是一个挑战。我们开发了 DeepGO-SE，这是一种使用预训练的大型语言模型根据蛋白质序列预测 GO 功能的方法。 DeepGO-SE 生成多个 GO 近似模型，神经网络预测这些近似模型中有关蛋白质功能的陈述的真值。我们汇总了多个模型的真值，以便 DeepGO-SE 在预测蛋白质功能时近似语义蕴涵。我们使用多个基准测试表明，与最先进的方法相比，该方法有效地利用了 GO 中的背景知识并改进了蛋白质功能预测。