ProteinNet: a standardized data set for machine learning of protein structure

By Mohammed Alquraishi. 2019

157 citations recorded by the source corpus.

Rapid progress in deep learning has spurred its application to bioinformatics problems including protein structure prediction and design. In classic machine learning problems like computer vision, progress has been driven by standardized data sets that facilitate fair assessment of new methods and lower the barrier to entry for non-domain experts. While data sets of protein sequence and structure exist, they lack certain components critical for machine learning, including high-quality multiple sequence alignments and insulated training/validation splits that account for deep but only weakly detectable homology across protein space. We created the ProteinNet series of data sets to provide a standardized mechanism for training and assessing data-driven models of protein sequence-structure relationships. ProteinNet integrates sequence, structure, and evolutionary information in programmatically accessible file formats tailored for machine learning frameworks. Multiple sequence alignments of all structurally characterized proteins were created using substantial high-performance computing resources. Standardized data splits were also generated to emulate the difficulty of past CASP (Critical Assessment of protein Structure Prediction) experiments by resetting protein sequence and structure space to the historical states that preceded six prior CASPs. Utilizing sensitive evolution-based distance metrics to segregate distantly related proteins, we have additionally created validation sets distinct from the official CASP sets that faithfully mimic their difficulty. ProteinNet represents a comprehensive and accessible resource for training and assessing machine-learned models of protein structure.

Subfields: Materials Chemistry, Molecular Biology, Evaluation & Benchmarks, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Materials Chemistry

In Molecular Biology

In Evaluation & Benchmarks

In Computational Biology & Bioinformatics

Citations in the corpus

BERTology Meets Biology: Interpreting Attention in Protein Language Models (2020, methodology): “We used two protein sequence datasets from the TAPE repository for the analysis: the ProteinNet dataset [3, 10, 26, 52] and the Secondary Structure dataset [10, 38, 52, 60].”
Modeling aspects of the language of life through transfer-learning protein sequences (2019, background): “0, we could not quite establish the validity of this assumption (analyzing pre-computed alignments from ProteinNet [59] revealed no clear relation of the type: more evolutionary information leads to better prediction).”
End-to-end differentiable learning of protein structure (2018, methodology): “To assess RGNs we therefore sought to recreate the conditions of past CASPs by assembling the ProteinNet datasets (Mohammed AlQuraishi, 2018).”
ProtTrans: Towards Cracking the Language of Life’s Code Through Self-Supervised Deep Learning and High Performance Computing (2020, background): “A comprehensive rigorous realization of data sets following this perspective has recently been published [85].”
Low-N protein engineering with data-efficient deep learning (2020, methodology): “In our examination of the mutational patterns in proposed and successful designs we began by gathering high-quality Position-Specific Scoring Matrices from the ProteinNet database 84 for both avGFP (PDB: 2WUR) and TEM-1 β-lactamase structure (PDB: 1ZG4).”
Single-sequence protein structure prediction using language models and deep learning (2022, methodology): “RGN2 training was performed using both the ProteinNet12 dataset 40 and a smaller dataset comprised solely of single protein domains derived from the ASTRAL SCOPe dataset (version 1.75) 41 .”
Single-sequence protein structure prediction using language models from deep learning (2021, methodology): “RGN2 training was performed using both the ProteinNet12 dataset 38 and a smaller dataset comprised solely of single protein domains derived from the ASTRAL SCOPe dataset (v1.”
Universal Transforming Geometric Network (2019, methodology): “Furthermore, a fully refined model may take months to train (AlQuraishi (2019a)).”
Profile Prediction: An Alignment-Based Pre-Training Task for Protein Sequence Models (2020, background): “, 2019), contact prediction (AlQuraishi, 2019), remote homology detection (Hou et al.”
Protein language model embeddings for fast, accurate, alignment-free protein structure prediction (2021, methodology): “We obtained 77,864 high-resolution experimental three-dimensional (3D) structures from ProteinNet12 (AlQuraishi, 2019) compiled from the PDB (Burley et al., 2017) before the CASP12 submission deadline (Moult et al., 2018) thereby replicating the CASP12 conditions.”
FLIP: Benchmark tasks in fitness landscape inference for proteins (2021, background): “The Critical Assessment of Protein Structure Prediction (CASP) [22], and retrospective protein training datasets from previous CASP competitions [23] have lowered the barrier to entry for new research teams and provided a clear account of progress over the last three decades [24].”
ProtTrans: Towards Cracking the Language of Life’s Code Through Self-Supervised Learning (2021, background): “Self-supervised pre-training combined with transfer-learning tapped into the gold-mines of unlabeled data opening the door for completely novel perspectives (and solutions) on existing problems.”