Bibliographic record
Abstract
Abstract Background The advent of protein embeddings has revolutionized bioinformatics by providing contextual representations that capture functional and evolutionary patterns. They have become, alongside sequence alignments, the cornerstone of bioinformatics. While embeddings cannot replace alignments, they can greatly help improving their quality. Our goal is replacing the BLOSUM matrices, the decades-old standard scoring system for protein alignment, with an embedding-based scoring method. Results We introduce a new scoring function and algorithm for local alignment of protein sequences, we offer a new comprehensive framework for evaluating local alignments. The score between two residues is given by the cosine similarity of their Ankh-embedding vectors and the algorithm uses dynamic programming with affine penalty. For the evaluation, we built multiple datasets, using both natural and inserted sequences, from the Conserved Domain Database, BAL-iBASE, and GPCRdb, designed a new algorithm for local alignment extraction, localization and quality evaluation, and employed five distance metrics to evaluate the similarity with the true alignment. We performed nearly one and a half million tests to compare the new algorithm with the best BLOSUM matrices, specialized GPCRtm matrices, and top programs, such as PEbA, ProtT5-score, DEDAL, vcMSA and pLM-BLAST. Regarding the protein embedding models, Ankh not only surpasses the best combination of ProtT5 and ESM2, but appears to better understand the “language” of proteins, as it behaves much better on natural sequences compared to artificial ones obtained by inserting domains in random protein sequences. Also, while ProtT5 and ESM2 combine to produce better results, Ankh does not combine well with other embeddings. Conclusions The new Ankh-score-based program is vastly superior to the BLOSUM matrices and clearly superior to all existing methods. New light shed on the protein embeddings can guide future improvements. In order to facilitate the use of the new method and protocol, they are freely available as a web server at e-score.csd.uwo.ca and as source code at github.com/lucian-ilie/E-score .
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 | 0.002 | 0.013 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.004 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
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".