Use AFSample2 Online
Commercially Available AFSample2 No-Code Web Server
AFSample2: Sample Alternative Protein Conformations with AlphaFold2 MSA Masking
AFsample2 is a method that increases the conformational diversity of AlphaFold2 (AF2) predictions by randomly masking columns of the multiple sequence alignment (MSA). Weakening the co-evolutionary signal lets AF2 sample alternative states, such as open and closed or inward- and outward-facing conformations, without degrading the models of the preferred state. The authors also propose a reference-free way to identify which predicted states are distinct.
AF2 typically returns one dominant conformation, yet many proteins, including transporters and enzymes with open and closed forms, work by switching between states. Standard AF2 sampling, and variants that rely on dropout or MSA clustering, often miss the alternate state. AFsample2 addresses this by perturbing the input evidence rather than the network.
Property | Detail |
|---|---|
Core idea | Random replacement of MSA columns with "X" to reduce co-evolutionary constraints |
Default masking | 15% of MSA columns (msa_rand_fraction) |
Default sampling | 1000 samples per target, all ten AF2 weight sets |
Base pipeline | AlphaFold v2.3.1, no templates |
Evaluation sets | 23 open/closed proteins (OC23) and a membrane transporter set |
How AFsample2 Generates Diverse Protein Conformations
Build the MSA: sequence databases are searched with the standard AF2 data pipeline (HHblits and Jackhmmer against UniRef90, BFD, Uniclust30 and MGnify). Templates are not used.
Mask random columns: a fraction of MSA columns is replaced with "X", set by the msa_rand_fraction parameter, with 15% as the default.
Run inference: each model run uses a uniquely masked MSA, across all ten AF2 weight sets.
Identify states: with reference structures the pipeline uses clustering, confidence and extremity selection. Without references it screens by confidence and then picks the model furthest from the most confident one.
Benchmark Results: Open-Closed Proteins and Membrane Transporters
On the OC23 set, mean best alternate-state TM-score rose from 0.795 with no masking to 0.878 at 15% masking, while the closed-state score changed only from 0.89 to 0.90. Performance dropped beyond about 30% masking.
Measure | Reported result |
|---|---|
OC23 targets with improved alternate state | 17 of 23 |
Both states at TM-score above 0.8 | 78.3% of targets, versus 47.8% for AFvanilla, AFdropout and AFcluster |
Intermediate conformations | About 70% more than AFvanilla |
Transporters with improved alternate state | 12 of 16 |
Transporter success at TM-score above 0.6 | 90% of targets versus 50% for other methods |
Ensemble Quality and Calibration
The fluctuation of AFsample2 ensembles correlated better with structural differences between states than the baselines for 19 of 23 targets, and the ensembles fell in the expected 2 to 5 calibration range for 16 of 23 targets.
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 AFSample2 on Tamarind Bio
Transporter mechanisms: sample inward- and outward-facing states of membrane transporters.
Open and closed states: model both conformations of proteins whose function depends on domain motion.
Intermediate states: generate in-between conformations for studying transitions, with the caveat that their on-path accuracy is not yet shown.
Fold-switching proteins: probe alternate folds, as the authors did with a modified S6 ribosomal protein.
How to Use AFSample2 on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select AFSample2.
Provide the sequence: enter the monomer protein sequence, since the method was tested on monomers.
Set masking: keep the 15% default for msa_rand_fraction unless exposed, or adjust it knowing that performance fell beyond about 30%.
Set sample count: the authors default to 1000 samples per target, if exposed.
Run the job: submit and wait for the ensemble to be generated.
Download outputs: collect the predicted structures and confidence scores.
Interpret states: compare models by TM-score or clustering, and inspect candidate alternate states manually.
Use downstream: take selected conformations into docking, design or simulation workflows.
Things to Keep in Mind
The method was tested only on monomers; multimeric complexes are left for future work.
Model confidence is a poor indicator of alternate states, which AF2 tends to rank lower.
Reference-free state selection can fail, so manual inspection is recommended.
Confidence declines as masking increases.
Fold-switch success was limited and may need more sampling.
Source: AFsample2 preprint on bioRxiv, https://www.biorxiv.org/content/10.1101/2024.05.28.596195v1 (2024). The pipeline builds on AlphaFold v2.3.1 (github.com/google-deepmind/alphafold). No dedicated AFsample2 code repository or license was stated on the pages read.