Figure 1.

Overview of ProLaTherm. After deriving sequence embeddings from the protein sequence, we retrieve protein language model embeddings by using the encoder part of ProtT5XLUniRef50. These are further average pooled along the sequence length to be processed by the head classifier. The head classifier consists of a fully connected layer with rectified linear activation, followed by batch normalization and a fully connected output layer. We further apply dropout before the average pooling and output layer.