Title:A Comparison of Mutual Information, Linear Models and Deep Learning
Networks for Protein Secondary Structure Prediction
Volume: 18
Issue: 8
Author(s): Saida Saad Mohamed Mahmoud, Beatrice Portelli, Giovanni D'Agostino, Gianluca Pollastri, Giuseppe Serra*Federico Fogolari*
Affiliation:
- Department of Mathematics, Computer Science and Physics, University of Udine, Udine, Italy
- Department of Mathematics, Computer Science and Physics, University of Udine, Udine, Italy
Keywords:
Secondary structure prediction, single sequence, mutual information, linear model, deep learning, neuralnetwork, LSTM, BERT.
Abstract:
Background: Over the last several decades, predicting protein structures from amino acid sequences
has been a core task in bioinformatics. Nowadays, the most successful methods employ multiple
sequence alignments and can predict the structure with excellent performance. These predictions
take advantage of all the amino acids at a given position and their frequencies. However, the effect of
single amino acid substitutions in a specific protein tends to be hidden by the alignment profile. For this
reason, single-sequence-based predictions attract interest even after accurate multiple-alignment methods
have become available: the use of single sequences ensures that the effects of substitution are not
confounded by homologous sequences.
Objective: This work aims at understanding how the single-sequence secondary structure prediction of a
residue is influenced by the surrounding ones. We aim at understanding how different prediction methods
use single-sequence information to predict the structure.
Methods: We compare mutual information, the coefficients of two linear models, and three deep learning
networks. For the deep learning algorithms, we use the DeepLIFT analysis to assess the effect of
each residue at each position in the prediction.
Results: Mutual information and linear models quantify direct effects, whereas DeepLIFT applied on
deep learning networks quantifies both direct and indirect effects.
Conclusion: Our analysis shows how different network architectures use the information of single protein
sequences and highlights their differences with respect to linear models. In particular, the deep
learning implementations take into account context and single position information differently, with the
best results obtained using the BERT architecture.