AcrTransAct: Pre-trained Protein Transformer Models for the Detection of Type I Anti-CRISPR Activities
Bibliographic record
Abstract
Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) and CRISPR-associated (Cas) proteins serve as a formidable defense mechanism for bacteria against foreign DNA; on the other hand, some bacteriophages (phages) and other mobile genetic elements have evolved anti-CRISPR (Acr) proteins to counteract CRISPR-Cas systems and ensure their own survival. Because Acr proteins provide phages with a fitness advantage relative to the bacteria that they infect, accurately identifying Acr proteins that inhibit CRISPR-Cas systems has the potential to significantly and positively impact our ability to harness phages to fight antimicrobial resistance. However, Acr identification is, at present, laborious and involves costly experimental procedures. Existing computational tools for protein-protein interaction (PPI) are not designed to predict complex inhibition, which could be the collective result of multiple PPIs. In this study, we developed a transformer-based deep neural network, AcrTransAct, to predict the probability of Acr-mediated CRISPR-Cas inhibition. Our model comprises two main components: 1. a feature extraction module that incorporates a pre-trained Evolutionary Scale Modeling (ESM) protein transformer and the NetSurfP-3.0 secondary structure prediction system; 2. a classification module that consists of either a convolutional or recurrent neural network. We created an inhibition dataset compiled from two Acr databases, AcrHub [30], Anti-CRISPRdb [5], and several published works [9, 12, 18, 20]. The AcrTransAct model is trained and tested on this dataset. We achieved an accuracy of 95% and an F1 score of 0.95 in predicting the inhibition of I-C, I-E, and I-F CRISPR-Cas systems by Acrs in our dataset. A web application of AcrTransAct (https://acrtransact.usask.ca) is implemented with the best-performing models from this study to predict the probability of multiple CRISPR-Cas systems inhibited by a putative Acr protein. Our code and data are available here: https://github.com/USask-BINFO/AcrTransAct.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".