Navigating the amino acid sequence space between functional proteins using a deep learning framework
Abstract
Motivation
Shedding light on the relationships between protein se-quences and functions is a challenging task with many implications in protein evolution, diseases understanding, and protein design. Protein sequence / function space is however hard to comprehend due to its com-plexity. Generative models help to decipher complex systems thanks to their abilities to learn and recreate data specificity. Applied to protein sequences, they can point out relationships between protein positions and functions capture the sequence patterns associated with functions or navigate through uncharted area of molecular evolution.
Results
In this study, an unsupervised generative approach based on adversarial auto-encoder (AAE) is proposed to generate and explore new sequences with respect to their functions thanks to the prior distribution allowing a continuous exploration of the latent space. AAEs are tested on three protein families known for their multiple functions. Clustering re-sults on the encoded sequences from the latent space computed by AAEs display high level of homogeneity regarding the protein sequence func-tions. The study also reports and analyzes for the first time two sampling strategies based on latent space interpolation and latent space arithmetic to generate intermediate protein sequences sharing sequential and functional properties of original sequences issued from different families and functions. Generated sequences by interpolation between latent space data points demonstrate the ability of the AAE to generalize and to pro-duce meaningful biological sequences from an evolutionary uncharted area of the biological sequence space. Finally, 3D structure models generated by comparative modelling between different combinations of structures of different sub-families and of generated sequences from latent space or sub-family sequences point out to the ability of the latent space arithmetic to successfully transfer functional properties between sub-families. All in all this study confirms the ability of deep learning frameworks to model biological complexity and bring new tools to explore amino acid sequence and functional spaces.
Availability
Code and data used for this study are freely available at <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="https://github.com/T-B-F/aae4seq">https://github.com/T-B-F/aae4seq</ext-link>.
Contact
<email>tristan@bitardfeildel.fr</email>
Supplementary information
Supplementary data are available at online.
Related articles
Related articles are currently not available for this article.