The introduction of AlphaFold 21 has spurred a revolution in modelling the structure of proteins and their interactions, enabling a huge range of applications in protein modelling and design2-6. Here we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture that is capable of predicting the joint structure of complexes including proteins, nucleic acids, small molecules, ions and modified residues. The new AlphaFold model demonstrates substantially improved accuracy over many previous specialized tools: far greater accuracy for protein-ligand interactions compared with state-of-the-art docking tools, much higher accuracy for protein-nucleic acid interactions compared with nucleic-acid-specific predictors and substantially higher antibody-antigen prediction accuracy compared with AlphaFold-Multimer v.2.37,8. Together, these results show that high-accuracy modelling across biomolecular space is possible within a single unified deep-learning framework.

We describe the operation and improvement of AlphaFold, the system that was entered by the team AlphaFold2 to the "human" category in the 14th Critical Assessment of Protein Structure Prediction (CASP14). The AlphaFold system entered in CASP14 is entirely different to the one entered in CASP13. It used a novel end-to-end deep neural network trained to produce protein structures from amino acid sequence, multiple sequence alignments, and homologous proteins. In the assessors' ranking by summed z scores (>2.0), AlphaFold scored 244.0 compared to 90.8 by the next best group. The predictions made by AlphaFold had a median domain GDT_TS of 92.4; this is the first time that this level of average accuracy has been achieved during CASP, especially on the more difficult Free Modeling targets, and represents a significant improvement in the state of the art in protein structure prediction. We reported how AlphaFold was run as a human team during CASP14 and improved such that it now achieves an equivalent level of performance without intervention, opening the door to highly accurate large-scale structure prediction.

Proteins are essential to life, and understanding their structure can facilitate a mechanistic understanding of their function. Through an enormous experimental effort 1–4 , the structures of around 100,000 unique proteins have been determined 5 , but this represents a small fraction of the billions of known protein sequences 6,7 . 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 years 9 . Despite recent progress 10–14 , 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.

While the vast majority of well-structured single protein chains can now be predicted to high accuracy due to the recent AlphaFold [1] model, the prediction of multi-chain protein complexes remains a challenge in many cases. In this work, we demonstrate that an AlphaFold model trained specifically for multimeric inputs of known stoichiometry, which we call AlphaFold-Multimer, significantly increases accuracy of predicted multimeric interfaces over input-adapted single-chain AlphaFold while maintaining high intra-chain accuracy. On a benchmark dataset of 17 heterodimer proteins without templates (introduced in [2]) we achieve at least medium accuracy (DockQ [3] ≥ 0.49) on 13 targets and high accuracy (DockQ ≥ 0.8) on 7 targets, compared to 9 targets of at least medium accuracy and 4 of high accuracy for the previous state of the art system (an AlphaFold-based system from [2]). We also predict structures for a large dataset of 4,446 recent protein complexes, from which we score all non-redundant interfaces with low template identity. For heteromeric interfaces we successfully predict the interface (DockQ ≥ 0.23) in 70% of cases, and produce high accuracy predictions (DockQ ≥ 0.8) in 26% of cases, an improvement of +27 and +14 percentage points over the flexible linker modification of AlphaFold [4] respectively. For homomeric inter-faces we successfully predict the interface in 72% of cases, and produce high accuracy predictions in 36% of cases, an improvement of +8 and +7 percentage points respectively.

Protein structures can provide invaluable information, both for reasoning about biological processes and for enabling interventions such as structure-based drug development or targeted mutagenesis. After decades of effort, 17% of the total residues in human protein sequences are covered by an experimentally determined structure 1 . Here we markedly expand the structural coverage of the proteome by applying the state-of-the-art machine learning method, AlphaFold 2 , at a scale that covers almost the entire human proteome (98.5% of human proteins). The resulting dataset covers 58% of residues with a confident prediction, of which a subset (36% of all residues) have very high confidence. We introduce several metrics developed by building on the AlphaFold model and use them to interpret the dataset, identifying strong multi-domain predictions as well as regions that are likely to be disordered. Finally, we provide some case studies to illustrate how high-quality predictions could be used to generate biological hypotheses. We are making our predictions freely available to the community and anticipate that routine large-scale and high-accuracy structure prediction will become an important tool that will allow new questions to be addressed from a structural perspective.

We present a method for combining multi-agent communication and traditional data-driven approaches to natural language learning, with an end goal of teaching agents to communicate with humans in natural language. Our starting point is a language model that has been trained on generic, not task-specific language data. We then place this model in a multi-agent self-play environment that generates task-specific rewards used to adapt or modulate the model, turning it into a task-conditional language model. We introduce a new way for combining the two types of learning based on the idea of reranking language model samples, and show that this method outperforms others in communicating with humans in a visual referential communication task. Finally, we present a taxonomy of different types of language drift that can occur alongside a set of measures to detect them.

We present the Compressive Transformer, an attentive sequence model which compresses past memories for long-range sequence learning. We find the Compressive Transformer obtains state-of-the-art language modelling results in the WikiText-103 and Enwik8 benchmarks, achieving 17.1 ppl and 0.97 bpc respectively. We also find it can model high-frequency speech effectively and can be used as a memory mechanism for RL, demonstrated on an object matching task. To promote the domain of long-range sequence learning, we propose a new open-vocabulary language modelling benchmark derived from books, PG-19.

Topic modeling is usually used to model words in documents by probabilistic mixtures of topics. We generalize this setup and consider arbitrary features of the positions in a corpus, e.g. “contains a word”, “belongs to a sentence”, “has a word in the local context”, “is labeled with a POS-tag”, etc. We build sparse probabilistic embeddings for positions and derive embeddings for the features by averaging of those. Importantly, we interpret the EM-algorithm as an iterative process of intersection and averaging steps that reestimate position and feature embeddings respectively. With this approach, we get several insights. First, we argue that a sentence should not be represented as an average of its words. While each word is a mixture of multiple senses, each word occurrence refers typically to just one specific sense. So in our approach, we obtain sentence embeddings by averaging position embeddings from the E-step. Second, we show that Biterm Topic Model (Yan et al. [11]) and Word Network Topic Model (Zuo et al. [12]) are equivalent with the only difference of tying word and context embeddings. We further extend these models by adjusting representation of each sliding window with a few iterations of EM-algorithm. Finally, we aim at consistent embeddings for hierarchical entities, e.g. for word-sentence-document structure. We discuss two alternative schemes of training and generalize to the case where the middle level of the hierarchy is unknown. It provides a unified formulation for topic segmentation and word sense disambiguation tasks.

Previous research on word embeddings has shown that sparse representations, which can be either learned on top of existing dense embeddings or obtained through model constraints during training time, have the benefit of increased interpretability properties: to some degree, each dimension can be understood by a human and associated with a recognizable feature in the data. In this paper, we transfer this idea to sentence embeddings and explore several approaches to obtain a sparse representation. We further introduce a novel, quantitative and automated evaluation metric for sentence embedding interpretability, based on topic coherence methods. We observe an increase in interpretability compared to dense models, on a dataset of movie dialogs and on the scene descriptions from the MS COCO dataset.

We consider probabilistic topic models and more recent word embedding techniques from a perspective of learning hidden semantic representations. Inspired by a striking similarity of the two approaches, we merge them and learn probabilistic embeddings with online EM-algorithm on word co-occurrence data. The resulting embeddings perform on par with Skip-Gram Negative Sampling (SGNS) on word similarity tasks and benefit in the interpretability of the components. Next, we learn probabilistic document embeddings that outperform paragraph2vec on a document similarity task and require less memory and time for training. Finally, we employ multimodal Additive Regularization of Topic Models (ARTM) to obtain a high sparsity and learn embeddings for other modalities, such as timestamps and categories. We observe further improvement of word similarity performance and meaningful inter-modality similarities.

Probabilistic topic modeling of text collections is a powerful tool for statistical text analysis. Determining the optimal number of topics remains a challenging problem in topic modeling. We propose a simple entropy regularization for topic selection in terms of Additive Regularization of Topic Models (ARTM), a multicriteria approach for combining regularizers. The entropy regularization gradually eliminates insignificant and linearly dependent topics. This process converges to the correct value on semi-real data. On real text collections it can be combined with sparsing, smoothing and decorrelation regularizers to produce a sequence of models with different numbers of well interpretable topics.

This paper considers a restriction to non-negative matrix factorization in which at least one matrix factor is stochastic. That is, the elements of the matrix factors are non-negative and the columns of one matrix factor sum to 1. This restriction includes topic models, a popular method for analyzing unstructured data. It also includes a method for storing and finding pictures. The paper presents necessary and sufficient conditions on the observed data such that the factorization is unique. In addition, the paper characterizes natural bounds on the parameters for any observed data and presents a consistent least squares estimator. The results are illustrated using a topic model analysis of PhD abstracts in economics and the problem of storing and retrieving a set of pictures of faces.

In this paper we introduce a generalized learning algorithm for probabilistic topic models (PTM). Many known and new algorithms for PLSA, LDA, and SWB models can be obtained as its special cases by choosing a subset of the following "options": regularization, sampling, update frequency, sparsing and robustness. We show that a robust topic model, which distinguishes specific, background and topic terms, doesn't need Dirichlet regularization and provides controllably sparse solution.

We propose a generalized probabilistic topic model of text corpora which can incorporate heuristics of Bayesian regularization, sampling, frequent parameters update, and robustness in any combinations. Wellknown models PLSA, LDA, CVB0, SWB, and many others can be considered as special cases of the proposed broad family of models. We propose the robust PLSA model and show that it is more sparse and performs better that regularized models like LDA.