Use COMPSS Protein Metrics Online
Commercially Available COMPSS Protein Metrics No-Code Web Server
COMPSS: Computational Metrics to Filter and Select Active Generated Enzyme Sequences
COMPSS (computational metrics for protein sequence selection) is a composite filter that scores generated enzyme sequences so that more of the ones you test in the lab are active. It was developed in work by Sean R. Johnson, Xiaozhi Fu, Sandra Viknander, Clara Goldin, Sarah Monaco, Aleksej Zelezniak, Kevin K. Yang and colleagues, published in Nature Biotechnology. The paper reports that over three rounds of experiments the filter raised experimental success rates by 50 to 150 percent.
Generative models such as ancestral sequence reconstruction, generative adversarial networks and protein language models can propose huge numbers of enzyme sequences, but many of them are not functional. Testing them one by one is costly, so researchers need computational metrics that predict which sequences will work before synthesis.
The authors evaluated 20 computational metrics on over 500 natural and generated sequences, with 70 to 90 percent identity to the nearest natural sequence, and used the results to build COMPSS.
Attribute | Detail |
|---|---|
Purpose | Select likely-functional generated enzyme sequences |
Metrics evaluated | 20 alignment-based, alignment-free and structure-based metrics |
Model enzyme families | Malate dehydrogenase (MDH) and copper superoxide dismutase (CuSOD) |
Generative models tested | Ancestral sequence reconstruction, ProteinGAN and ESM-MSA |
Experimental readout | Expression in E. coli and in vitro activity assays |
How the COMPSS Filter Selects Functional Enzyme Sequences
The metrics fall into three groups, which the filter combines.
Alignment-based: sequence identity, BLOSUM62, PFASUM15, phmmer and ESM-MSA scores.
Alignment-free: CARP-640M, ESM-1v and net charge.
Structure-based: AlphaFold2 pLDDT, Rosetta, solvent accessible surface area, ProteinMPNN, ESM-IF and MIF-ST.
The filtering steps
Sequence quality checks: for example, the sequence starts with methionine and has no long repeats or transmembrane domains.
Language-model cutoff: an ESM-1v threshold set at the 10th percentile of natural sequences.
Structure prediction: AlphaFold2 models the surviving sequences.
Inverse-folding ranking: ProteinMPNN scores rank the remaining candidates.
Experimental Results: How Much COMPSS Improves Enzyme Activity Rates
Result | Reported value |
|---|---|
Active sequences in round 1 (naive generation) | 19% |
COMPSS-selected sequences active in round 3 | 74% |
Improvement over controls in round 3 | 77% higher success rate |
ESM-MSA designs active in round 3 | 94% of CuSOD (17/18) and 100% of MDH |
Active selected sequences within 10-fold of wild-type activity | 83% (44/53) |
Inverse-folding metrics had the best average AUC-ROC (0.72) in round 2, and sequence identity to natural proteins did not predict activity. The authors also checked published datasets from six additional families, and in five of them filter-passing sequences had a higher functional fraction.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with COMPSS Protein Metrics on Tamarind Bio
Prioritizing generated enzymes: rank sequences from a generative model before ordering them for expression.
Comparing generative models: compare ancestral reconstruction, GAN and language-model outputs with the same metrics.
Removing likely failures: flag sequences with signal peptides, transmembrane domains or other quality problems.
Library design: shrink large candidate libraries to a smaller, higher-quality set for testing.
How to Use COMPSS Protein Metrics on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select COMPSS Protein Metrics.
Provide sequences: upload the candidate enzyme sequences as protein sequences, for example in FASTA format, if exposed.
Provide a natural reference set: supply natural homologs for alignment-based metrics and the ESM-1v percentile cutoff, if exposed.
Choose metrics: select the metrics or filter to apply, if exposed.
Run the job: launch it and monitor progress from your dashboard.
Download outputs: retrieve per-sequence scores and pass or fail calls.
Interpret the scores: prioritize sequences that pass the quality checks and rank highly on the structure-based metrics.
Use results downstream: select sequences for synthesis and expression, and validate with activity assays.
Things to Keep in Mind
No single metric was superior, and AUC-ROC values were moderate.
Only two enzyme families were tested in depth, so the authors caution against applying COMPSS blindly to other families.
AlphaFold2 pLDDT did not consistently predict activity.
Performance on ProGen lysozymes was weak, suggesting fine-tuning may be needed.
Expression and activity were tested only in E. coli with in vitro assays.
The strict ESM-1v cutoff in round 3 introduced phylogenetic bias, especially for MDH.
Source: Sean R. Johnson, Xiaozhi Fu, Sandra Viknander, Clara Goldin, Sarah Monaco, Aleksej Zelezniak et al., Computational scoring and experimental evaluation of enzymes generated by neural networks, Nature Biotechnology 43(3):396-405 (published online 2024). DOI 10.1038/s41587-024-02214-2. Paper: https://www.nature.com/articles/s41587-024-02214-2