PHA Synthase Sequences Generated by Conditional Variational Autoencoder

Approximately 10,000 novel Polyhydroxyalkanoate synthase protein sequences were generated using a conditional variational autoencoder trained on native sequences from Uniprot. The dataset includes 16 sequences selected for in vivo validation, with two confirmed active variants producing poly(hydroxybutyrate) in yeast. The data was authored by Tuula Tenkanen and last updated on March 19, 2026.

Use Cases

• Training generative models for protein sequence design based on native enzyme sequences
• Evaluating novel enzyme activity based on conserved residues and structural features
• Studying the relationship between amino acid substitutions and enzyme function

Strengths

• Contains approximately 10,000 generated protein sequences
• Includes 16 sequences selected for validation with specific selection criteria
• Two novel enzymes were confirmed active with 87 and 98 amino acid substitutions

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• The dataset is 4.9 MB, which is a small scale

Provenance

Source

figshare

Collection Method

Generated using a conditional variational autoencoder trained on native protein sequences from Uniprot

Freshness

Last updated 2026-03-19 17:34:29