Research map / Probabilistic and scientific modeling / Molecules and proteins
Nested birth-death processes are competitive with neural networks as time-dependent models of protein evolution
Annabel Large, Ian Holmes
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Basic neural head with Neural TKF (TKF92-based) head (ablation) | 6-block Transformer embedders · Pfam pairwise alignments test | ECE: 5.6992 -> 5.5169 | Improved |
| replace Basic neural head with Neural TKF head (ablation) | LSTM embedders · Pfam pairwise alignments test | ECE: 5.6877 -> 5.5340 | Improved |
| replace Basic neural head with Neural TKF head (ablation) | Residual CNN embedders · Pfam pairwise alignments test | ECE: 5.7455 -> 5.6258 | Improved |
| replace Basic neural head with Neural TKF head (ablation) | 1-block Transformer embedders · Pfam pairwise alignments test | ECE: 5.8817 -> 5.6983 | Improved |
| scale Transformer from 1 block to 6 blocks (hyperparameter sweep, Neural TKF) (ablation) | Neural TKF · Pfam pairwise alignments test | ECE: 5.6983 -> 5.5169 | Improved |
| tune Transformer hyperparameters, tied to Basic neural head (ablation) | Basic neural · Pfam pairwise alignments test | ECE: 5.6992 (6-block) vs 5.6877 (LSTM) | No clear effect |
| use TKF92 instead of TKF91 as indel model (ablation) | F81 indel model · Pfam pairwise alignments test | ECE: 7.7942 -> 7.1398 | Improved |
| use H20 (renormalized ODE) instead of TKF92 (ablation) | indel model · Pfam pairwise alignments test | ECE: 7.1398 -> 7.1530 | Hurt |
| add 30 site classes to TKF92+F81 (ablation) | TKF92+F81 · Pfam pairwise alignments test | ECE: 7.1398 -> 6.1238 | Improved |
| add 30 fragment classes on top of site classes (ablation) | mixture of site classes · Pfam pairwise alignments test | ECE: 6.1238 -> 5.9602 | Improved |
| add 10 domain classes (MixDom) on top of fragment and site classes (ablation) | mixture of fragment classes · Pfam pairwise alignments test | ECE: 5.9602 -> 5.7013 | Improved |
| increase number of mixture components within a single level (ablation) | TKF-based mixtures · Pfam pairwise alignments (AIC gain) | delta AIC | No clear effect |
Closest papers by meaning
- Ultrafast classical phylogenetic method beats large protein language models on variant effect prediction (NeurIPS 2024 · similarity 0.87)
- Deep Generative Models for Phylogenetic Inference with Complex Evolutionary Processes (NeurIPS 2026 · similarity 0.87)
- PhyloGen: Language Model-Enhanced Phylogenetic Inference via Graph Structure Generation (NeurIPS 2024 · similarity 0.84)
- Evolution-Inspired Loss Functions for Protein Representation Learning (ICML 2024 · similarity 0.84)
- DePLM: Denoising Protein Language Models for Property Optimization (NeurIPS 2024 · similarity 0.84)
- LEMON-ZEST: Evolution-Informed Tokenization for Efficient Protein Language Modeling (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
