论文标题
基于变压器的分子编码财产预测
Transformer Based Molecule Encoding for Property Prediction
论文作者
论文摘要
分子性质预测的神经方法需要有效地编码结构和性质关系。使用图算法的最新工作显示，潜在分子编码空间中的概括有限。我们构建了一个基于变压器的分子编码器和属性预测网络，具有新颖的输入功能，其性能明显优于现有方法。我们将模型适应半监督的学习，以进一步在通常可用的有限实验数据上表现出色。
Neural methods of molecule property prediction require efficient encoding of structure and property relationship to be accurate. Recent work using graph algorithms shows limited generalization in the latent molecule encoding space. We build a Transformer-based molecule encoder and property predictor network with novel input featurization that performs significantly better than existing methods. We adapt our model to semi-supervised learning to further perform well on the limited experimental data usually available in practice.
