Proteins are essential to life, and understanding their structure can facilitate a mechanistic understanding of their function. Through an enormous experimental effort1–4, the structures of around 100,000 unique proteins have been determined5, but this represents a small fraction of the billions of known protein sequences6,7. Structural coverage is bottlenecked by the months to years of painstaking effort required to determine a single protein structure. Accurate computational approaches are needed to address this gap and to enable large-scale structural bioinformatics. Predicting the three-dimensional structure that a protein will adopt based solely on its amino acid sequence—the structure prediction component of the ‘protein folding problem’8—has been an important open research problem for more than 50 years9. Despite recent progress10–14, existing methods fall far short of atomic accuracy, especially when no homologous structure is available. Here we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known. We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14)15, demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods. Underpinning the latest version of AlphaFold is a novel machine learning approach that incorporates physical and biological knowledge about protein structure, leveraging multi-sequence alignments, into the design of the deep learning algorithm. AlphaFold predicts protein structures with an accuracy competitive with experimental structures in the majority of cases using a novel deep learning architecture.

/pdf/highly-accurate-protein-structure-prediction-with-alphafold-2xjbzfwu3n.pdf

Highly accurate protein structure prediction with AlphaFold

A morphable model for the synthesis of 3D faces

Inspired by progress in unsupervised representation learning for natural language, we examine whether similar models can learn useful representations for images. We train a sequence Transformer to auto-regressively predict pixels, without incorporating knowledge of the 2D input structure. Despite training on low-resolution ImageNet without labels, we find that a GPT-2 scale model learns strong image representations as measured by linear probing, fine-tuning, and low-data classification. On CIFAR-10, we achieve 96.3% accuracy with a linear probe, outperforming a supervised Wide ResNet, and 99.0% accuracy with full finetuning, matching the top supervised pre-trained models. An even larger model trained on a mixture of ImageNet and web images is competitive with self-supervised benchmarks on ImageNet, achieving 72.0% top-1 accuracy on a linear probe of our features.

/pdf/generative-pretraining-from-pixels-nta4i7dig7.pdf

Generative Pretraining From Pixels

In the field of artificial intelligence, a combination of scale in data and model capacity enabled by unsupervised learning has led to major advances in representation learning and statistical generation. In biology, the anticipated growth of sequencing promises unprecedented data on natural sequence diversity. Learning the natural distribution of evolutionary protein sequence variation is a logical step toward predictive and generative modeling for biology. To this end we use unsupervised learning to train a deep contextual language model on 86 billion amino acids across 250 million sequences spanning evolutionary diversity. The resulting model maps raw sequences to representations of biological properties without labels or prior domain knowledge. The learned representation space organizes sequences at multiple levels of biological granularity from the biochemical to proteomic levels. Learning recovers information about protein structure: secondary structure and residue-residue contacts can be extracted by linear projections from learned representations. With small amounts of labeled data, the ability to identify tertiary contacts is further improved. Learning on full sequence diversity rather than individual protein families increases recoverable information about secondary structure. We show the networks generalize by adapting them to variant activity prediction from sequences only, with results that are comparable to a state-of-the-art variant predictor that uses evolutionary and structurally derived features.

https://www.biorxiv.org/content/biorxiv/early/2019/05/29/622803.full.pdf

Biological Structure and Function Emerge from Scaling Unsupervised Learning to 250 Million Protein Sequences

In the field of artificial intelligence, a combination of scale in data and model capacity enabled by unsupervised learning has led to major advances in representation learning and statistical generation In the life sciences, the anticipated growth of sequencing promises unprecedented data on natural sequence diversity Protein language modeling at the scale of evolution is a logical step toward predictive and generative artificial intelligence for biology To this end, we use unsupervised learning to train a deep contextual language model on 86 billion amino acids across 250 million protein sequences spanning evolutionary diversity The resulting model contains information about biological properties in its representations The representations are learned from sequence data alone The learned representation space has a multiscale organization reflecting structure from the level of biochemical properties of amino acids to remote homology of proteins Information about secondary and tertiary structure is encoded in the representations and can be identified by linear projections Representation learning produces features that generalize across a range of applications, enabling state-of-the-art supervised prediction of mutational effect and secondary structure and improving state-of-the-art features for long-range contact prediction

https://www.pnas.org/content/pnas/118/15/e2016239118.full.pdf

Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences

A probabilistic password model assigns a probability value to each string. Such models are useful for research into understanding what makes users choose more (or less) secure passwords, and for constructing password strength meters and password cracking utilities. Guess number graphs generated from password models are a widely used method in password research. In this paper, we show that probability-threshold graphs have important advantages over guess-number graphs. They are much faster to compute, and at the same time provide information beyond what is feasible in guess-number graphs. We also observe that research in password modeling can benefit from the extensive literature in statistical language modeling. We conduct a systematic evaluation of a large number of probabilistic password models, including Markov models using different normalization and smoothing methods, and found that, among other things, Markov models, when done correctly, perform significantly better than the Probabilistic Context-Free Grammar model proposed in Weir et al., which has been used as the state-of-the-art password model in recent research.

/pdf/a-study-of-probabilistic-password-models-4aogrhnqyv.pdf

A Study of Probabilistic Password Models

Momentum-based acceleration of stochastic gradient descent (SGD) is widely used in deep learning. We propose the quasi-hyperbolic momentum algorithm (QHM) as an extremely simple alteration of momentum SGD, averaging a plain SGD step with a momentum step. We describe numerous connections to and identities with other algorithms, and we characterize the set of two-state optimization algorithms that QHM can recover. Finally, we propose a QH variant of Adam called QHAdam, and we empirically demonstrate that our algorithms lead to significantly improved training in a variety of settings, including a new state-of-the-art result on WMT16 EN-DE. We hope that these empirical results, combined with the conceptual and practical simplicity of QHM and QHAdam, will spur interest from both practitioners and researchers. Code is immediately available.

Quasi-hyperbolic momentum and Adam for deep learning

The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning's capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy. However, many obstacles remain in the understanding of and usability of these promising approaches by the research community. Toward elucidating unresolved mysteries and facilitating future research, we propose ELF OpenGo, an open-source reimplementation of the AlphaZero algorithm. ELF OpenGo is the first open-source Go AI to convincingly demonstrate superhuman performance with a perfect (20:0) record against global top professionals. We apply ELF OpenGo to conduct extensive ablation studies, and to identify and analyze numerous interesting phenomena in both the model training and in the gameplay inference procedures. Our code, models, selfplay datasets, and auxiliary data are publicly available.

Jerry Ma

Papers

Biological Structure and Function Emerge from Scaling Unsupervised Learning to 250 Million Protein Sequences

Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences

A Study of Probabilistic Password Models

Quasi-hyperbolic momentum and Adam for deep learning

ELF OpenGo: An Analysis and Open Reimplementation of AlphaZero