SCOPe: improvements to the structural classification of proteins - extended database to facilitate variant interpretation and machine learning.

The Structural Classification of Proteins-extended (SCOPe, https://scop.berkeley.edu) knowledgebase aims to provide an accurate, detailed, and comprehensive description of the structural and evolutionary relationships amongst the majority of proteins of known structure, along with resources for analyzing the protein structures and their sequences. Structures from the PDB are divided into domains and classified using a combination of manual curation and highly precise automated methods. In the current release of SCOPe, 2.08, we have developed search and display tools for analysis of genetic variants we mapped to structures classified in SCOPe. In order to improve the utility of SCOPe to automated methods such as deep learning classifiers that rely on multiple alignment of sequences of homologous proteins, we have introduced new machine-parseable annotations that indicate aberrant structures as well as domains that are distinguished by a smaller repeat unit. We also classified structures from 74 of the largest Pfam families not previously classified in SCOPe, and we improved our algorithm to remove N- and C-terminal cloning, expression and purification sequences from SCOPe domains. SCOPe 2.08-stable classifies 106 976 PDB entries (about 60% of PDB entries).

FLIGHTED: Inferring Fitness Landscapes from Noisy High-Throughput Experimental Data.

Machine learning (ML) for protein design requires large protein fitness datasets generated by high-throughput experiments for training, fine-tuning, and benchmarking models. However, most models do not account for experimental noise inherent in these datasets, harming model performance and changing model rankings in benchmarking studies. Here we develop FLIGHTED, a Bayesian method of accounting for uncertainty by generating probabilistic fitness landscapes from noisy high-throughput experiments. We demonstrate how FLIGHTED can improve model performance on two categories of experiments: single-step selection assays, such as phage display and SELEX, and a novel high-throughput assay called DHARMA that ties activity to base editing. We then compare the performance of standard machine-learning models on fitness landscapes generated with and without FLIGHTED. Accounting for noise significantly improves model performance, especially of CNN architectures, and changes relative rankings on numerous common benchmarks. Based on our new benchmarking with FLIGHTED, data size, not model scale, currently appears to be limiting the performance of protein fitness models, and the choice of top model architecture matters more than the protein language model embedding. Collectively, our results indicate that FLIGHTED can be applied to any high-throughput assay and any machine learning model, making it straightforward for protein designers to account for experimental noise when modeling protein fitness.