Unsupervised Language Model Research Articles

Crystal Composition Transformer: Self-Learning Neural Language Model for Generative and Tinkering Design of Materials.

Self-supervised neural language models have recently achieved unprecedented success from natural language processing to learning the languages of biological sequences and organic molecules. These models have demonstrated superior performance in the generation, structure classification, and functional predictions for proteins and molecules with learned representations. However, most of the masking-based pre-trained language models are not designed for generative design, and their black-box nature makes it difficult to interpret their design logic. Here a Blank-filling Language Model for Materials (BLMM) Crystal Transformer is proposed, a neural network-based probabilistic generative model for generative and tinkering design of inorganic materials. The model is built on the blank-filling language model for text generation and has demonstrated unique advantages in learning the "materials grammars" together with high-quality generation, interpretability, and data efficiency. It can generate chemically valid materials compositions with as high as 89.7% charge neutrality and 84.8% balanced electronegativity, which are more than four and eight times higher compared to a pseudo-random sampling baseline. The probabilistic generation process of BLMM allows it to recommend materials tinkering operations based on learned materials chemistry, which makes it useful for materials doping. The model is applied to discover a set of new materials as validated using the Density Functional Theory (DFT) calculations. This work thus brings the unsupervised transformer language models based generative artificial intelligence to inorganic materials. A user-friendly web app for tinkering materials design has been developed and can be accessed freely at www.materialsatlas.org/blmtinker.

TULIP: A transformer-based unsupervised language model for interacting peptides and T cell receptors that generalizes to unseen epitopes

The accurate prediction of binding between T cell receptors (TCR) and their cognate epitopes is key to understanding the adaptive immune response and developing immunotherapies. Current methods face two significant limitations: the shortage of comprehensive high-quality data and the bias introduced by the selection of the negative training data commonly used in the supervised learning approaches. We propose a method, Transformer-based Unsupervised Language model for Interacting Peptides and T cell receptors (TULIP), that addresses both limitations by leveraging incomplete data and unsupervised learning and using the transformer architecture of language models. Our model is flexible and integrates all possible data sources, regardless of their quality or completeness. We demonstrate the existence of a bias introduced by the sampling procedure used in previous supervised approaches, emphasizing the need for an unsupervised approach. TULIP recognizes the specific TCRs binding an epitope, performing well on unseen epitopes. Our model outperforms state-of-the-art models and offers a promising direction for the development of more accurate TCR epitope recognition models.

Prediction of protein-ATP binding residues using multi-view feature learning via contextual-based co-attention network

Accurately predicting protein-ATP binding residues is critical for protein function annotation and drug discovery. Computational methods dedicated to the prediction of binding residues based on protein sequence information have exhibited notable advancements in predictive accuracy. Nevertheless, these methods continue to grapple with several formidable challenges, including limited means of extracting more discriminative features and inadequate algorithms for integrating protein and residue information. To address the problems, we propose ATP-Deep, a novel protein-ATP binding residues predictor. ATP-Deep harnesses the capabilities of unsupervised pre-trained language models and incorporates domain-specific evolutionary context information from homologous sequences. It further refines the embedding at the residue level through integration with corresponding protein-level information and employs a contextual-based co-attention mechanism to adeptly fuse multiple sources of features. The performance evaluation results on the benchmark datasets reveal that ATP-Deep achieves an AUC of 0.954 and 0.951, respectively, surpassing the performance of the state-of-the-art model. These findings underscore the effectiveness of assimilating protein-level information and deploying a contextual-based co-attention mechanism grounded in context to bolster the prediction performance of protein-ATP binding residues.

ULDNA: integrating unsupervised multi-source language models with LSTM-attention network for high-accuracy protein-DNA binding site prediction.

Efficient and accurate recognition of protein-DNA interactions is vital for understanding the molecular mechanisms of related biological processes and further guiding drug discovery. Although the current experimental protocols are the most precise way to determine protein-DNA binding sites, they tend to be labor-intensive and time-consuming. There is an immediate need to design efficient computational approaches for predicting DNA-binding sites. Here, we proposed ULDNA, a new deep-learning model, to deduce DNA-binding sites from protein sequences. This model leverages an LSTM-attention architecture, embedded with three unsupervised language models that are pre-trained on large-scale sequences from multiple database sources. To prove its effectiveness, ULDNA was tested on 229 protein chains with experimental annotation of DNA-binding sites. Results from computational experiments revealed that ULDNA significantly improves the accuracy of DNA-binding site prediction in comparison with 17 state-of-the-art methods. In-depth data analyses showed that the major strength of ULDNA stems from employing three transformer language models. Specifically, these language models capture complementary feature embeddings with evolution diversity, in which the complex DNA-binding patterns are buried. Meanwhile, the specially crafted LSTM-attention network effectively decodes evolution diversity-based embeddings as DNA-binding results at the residue level. Our findings demonstrated a new pipeline for predicting DNA-binding sites on a large scale with high accuracy from protein sequence alone.

Multiple sequence alignment-based RNA language model and its application to structural inference.

Compared with proteins, DNA and RNA are more difficult languages to interpret because four-letter coded DNA/RNA sequences have less information content than 20-letter coded protein sequences. While BERT (Bidirectional Encoder Representations from Transformers)-like language models have been developed for RNA, they are ineffective at capturing the evolutionary information from homologous sequences becauseunlike proteins, RNA sequences are less conserved. Here, we have developed an unsupervised multiple sequence alignment-based RNA language model (RNA-MSM) by utilizing homologous sequences from an automatic pipeline, RNAcmap, as it can provide significantly more homologous sequences than manually annotated Rfam. We demonstrate that the resulting unsupervised, two-dimensional attention maps and one-dimensional embeddings from RNA-MSM contain structural information. In fact, they can be directly mapped with high accuracy to 2D base pairing probabilities and 1D solvent accessibilities, respectively. Further fine-tuning led to significantly improved performance on these two downstream tasks compared with existing state-of-the-art techniques including SPOT-RNA2 and RNAsnap2. By comparison, RNA-FM, a BERT-based RNA language model, performs worse than one-hot encoding with its embedding in base pair and solvent-accessible surface area prediction. We anticipate that the pre-trained RNA-MSM model can be fine-tuned on many other tasks related to RNA structure and function.

Integrating unsupervised language model with multi-view multiple sequence alignments for high-accuracy inter-chain contact prediction

Accurate identification of inter-chain contacts in the protein complex is critical to determine the corresponding 3D structures and understand the biological functions. We proposed a new deep learning method, ICCPred, to deduce the inter-chain contacts from the amino acid sequences of the protein complex. This pipeline was built on the designed deep residual network architecture, integrating the pre-trained language model with three multiple sequence alignments (MSAs) from different biological views. Experimental results on 709 non-redundant benchmarking protein complexes showed that the proposed ICCPred significantly increased inter-chain contact prediction accuracy compared to the state-of-the-art approaches. Detailed data analyses showed that the significant advantage of ICCPred lies in the utilization of pre-trained transformer language models which can effectively extract the complementary co-evolution diversity from three MSAs. Meanwhile, the designed deep residual network enhances the correlation between the co-evolution diversity and the patterns of inter-chain contacts. These results demonstrated a new avenue for high-accuracy deep-learning inter-chain contact prediction that is applicable to large-scale protein-protein interaction annotations from sequence alone.

Fast and accurate protein intrinsic disorder prediction by using a pretrained language model.

Determining intrinsically disordered regions of proteins is essential for elucidating protein biological functions and the mechanisms of their associated diseases. As the gap between the number of experimentally determined protein structures and the number of protein sequences continues to grow exponentially, there is a need for developing an accurate and computationally efficient disorder predictor. However, current single-sequence-based methods are of low accuracy, while evolutionary profile-based methods are computationally intensive. Here, we proposed a fast and accurate protein disorder predictor LMDisorder that employed embedding generated by unsupervised pretrained language models as features. We showed that LMDisorder performs best in all single-sequence-based methods and is comparable or better than another language-model-based technique in four independent test sets, respectively. Furthermore, LMDisorder showed equivalent or even better performance than the state-of-the-art profile-based technique SPOT-Disorder2. In addition, the high computation efficiency of LMDisorder enabled proteome-scale analysis of human, showing that proteins with high predicted disorder content were associated with specific biological functions. The datasets, the source codes, and the trained model are available at https://github.com/biomed-AI/LMDisorder.

Text Generator using GPT2 Model

Abstract- GPT-2 is state of art algorithm transfer learning with respect to nlp task we can do it like text summarizationand many more applications. The text generation application programming interface is supported by a large-scale unsupervised language model capable of generating paragraphs of text This transformer-based language model, based on OpenAI's GPT-2 model, accepts a sentence or partial sentence as input and predicts subsequent text. GPT-2 uses input textto set the initial context for further text generation. The length of an input string can range from few words to a maximumsequence length of 1024 tokens. The longer an initial input, the more subject context is provided to a model. Generally, longer inputs produce a more coherent text output. It was specifically trained to guess the next word in sentences. GPT-2 isa massive model that contains a massive amount of compressed knowledge from various parts of the internet. It can be usedto forecast the likelihood of a sentence. The model learns an internal representation of the English language, which it can then use to extract features useful for subsequent tasks. We have used a tkinter which is a GUI to display the generated output. Keywords—Text generation, Graphical User Interface, gpt2 transformer,Open AI,tkinter,

SHINE: protein language model-based pathogenicity prediction for short inframe insertion and deletion variants.

Accurate variant pathogenicity predictions are important in genetic studies of human diseases. Inframe insertion and deletion variants (indels) alter protein sequence and length, but not as deleterious as frameshift indels. Inframe indel Interpretation is challenging due to limitations in the available number of known pathogenic variants for training. Existing prediction methods largely use manually encoded features including conservation, protein structure and function, and allele frequency to infer variant pathogenicity. Recent advances in deep learning modeling of protein sequences and structures provide an opportunity to improve the representation of salient features based on large numbers of protein sequences. We developed a new pathogenicity predictor for SHort Inframe iNsertion and dEletion (SHINE). SHINE uses pretrained protein language models to construct a latent representation of an indel and its protein context from protein sequences and multiple protein sequence alignments, and feeds the latent representation into supervised machine learning models for pathogenicity prediction. We curated training data from ClinVar and gnomAD, and created two test datasets from different sources. SHINE achieved better prediction performance than existing methods for both deletion and insertion variants in these two test datasets. Our work suggests that unsupervised protein language models can provide valuable information about proteins, and new methods based on these models can improve variant interpretation in genetic analyses.

Integrating unsupervised language model with triplet neural networks for protein gene ontology prediction.

Accurate identification of protein function is critical to elucidate life mechanisms and design new drugs. We proposed a novel deep-learning method, ATGO, to predict Gene Ontology (GO) attributes of proteins through a triplet neural-network architecture embedded with pre-trained language models from protein sequences. The method was systematically tested on 1068 non-redundant benchmarking proteins and 3328 targets from the third Critical Assessment of Protein Function Annotation (CAFA) challenge. Experimental results showed that ATGO achieved a significant increase of the GO prediction accuracy compared to the state-of-the-art approaches in all aspects of molecular function, biological process, and cellular component. Detailed data analyses showed that the major advantage of ATGO lies in the utilization of pre-trained transformer language models which can extract discriminative functional pattern from the feature embeddings. Meanwhile, the proposed triplet network helps enhance the association of functional similarity with feature similarity in the sequence embedding space. In addition, it was found that the combination of the network scores with the complementary homology-based inferences could further improve the accuracy of the predicted models. These results demonstrated a new avenue for high-accuracy deep-learning function prediction that is applicable to large-scale protein function annotations from sequence alone.

Large-scale chemical language representations capture molecular structure and properties

Models based on machine learning can enable accurate and fast molecular property predictions, which is of interest in drug discovery and material design. Various supervised machine learning models have demonstrated promising performance, but the vast chemical space and the limited availability of property labels make supervised learning challenging. Recently, unsupervised transformer-based language models pretrained on a large unlabelled corpus have produced state-of-the-art results in many downstream natural language processing tasks. Inspired by this development, we present molecular embeddings obtained by training an efficient transformer encoder model, MoLFormer, which uses rotary positional embeddings. This model employs a linear attention mechanism, coupled with highly distributed training, on SMILES sequences of 1.1 billion unlabelled molecules from the PubChem and ZINC datasets. We show that the learned molecular representation outperforms existing baselines, including supervised and self-supervised graph neural networks and language models, on several downstream tasks from ten benchmark datasets. They perform competitively on two others. Further analyses, specifically through the lens of attention, demonstrate that MoLFormer trained on chemical SMILES indeed learns the spatial relationships between atoms within a molecule. These results provide encouraging evidence that large-scale molecular language models can capture sufficient chemical and structural information to predict various distinct molecular properties, including quantum-chemical properties. Large language models have recently emerged with extraordinary capabilities, and these methods can be applied to model other kinds of sequence, such as string representations of molecules. Ross and colleagues have created a transformer-based model, trained on a large dataset of molecules, which provides good results on property prediction tasks.

ProtGPT2 is a deep unsupervised language model for protein design

Protein design aims to build novel proteins customized for specific purposes, thereby holding the potential to tackle many environmental and biomedical problems. Recent progress in Transformer-based architectures has enabled the implementation of language models capable of generating text with human-like capabilities. Here, motivated by this success, we describe ProtGPT2, a language model trained on the protein space that generates de novo protein sequences following the principles of natural ones. The generated proteins display natural amino acid propensities, while disorder predictions indicate that 88% of ProtGPT2-generated proteins are globular, in line with natural sequences. Sensitive sequence searches in protein databases show that ProtGPT2 sequences are distantly related to natural ones, and similarity networks further demonstrate that ProtGPT2 is sampling unexplored regions of protein space. AlphaFold prediction of ProtGPT2-sequences yields well-folded non-idealized structures with embodiments and large loops and reveals topologies not captured in current structure databases. ProtGPT2 generates sequences in a matter of seconds and is freely available.

Sentence Similarity Based on Contexts

AbstractExisting methods to measure sentence similarity are faced with two challenges: (1) labeled datasets are usually limited in size, making them insufficient to train supervised neural models; and (2) there is a training-test gap for unsupervised language modeling (LM) based models to compute semantic scores between sentences, since sentence-level semantics are not explicitly modeled at training. This results in inferior performances in this task. In this work, we propose a new framework to address these two issues. The proposed framework is based on the core idea that the meaning of a sentence should be defined by its contexts, and that sentence similarity can be measured by comparing the probabilities of generating two sentences given the same context. The proposed framework is able to generate high-quality, large-scale dataset with semantic similarity scores between two sentences in an unsupervised manner, with which the train-test gap can be largely bridged. Extensive experiments show that the proposed framework achieves significant performance boosts over existing baselines under both the supervised and unsupervised settings across different datasets.

Framework for Deep Learning-Based Language Models Using Multi-Task Learning in Natural Language Understanding: A Systematic Literature Review and Future Directions

Learning human languages is a difficult task for a computer. However, Deep Learning (DL) techniques have enhanced performance significantly for almost all-natural language processing (NLP) tasks. Unfortunately, these models cannot be generalized for all the NLP tasks with similar performance. NLU (Natural Language Understanding) is a subset of NLP including tasks, like machine translation, dialogue-based systems, natural language inference, text entailment, sentiment analysis, etc. The advancement in the field of NLU is the collective performance enhancement in all these tasks. Even though MTL (Multi-task Learning) was introduced before Deep Learning, it has gained significant attention in the past years. This paper aims to identify, investigate, and analyze various language models used in NLU and NLP to find directions for future research. The Systematic Literature Review (SLR) is prepared using the literature search guidelines proposed by Kitchenham and Charters on various language models between 2011 and 2021. This SLR points out that the unsupervised learning method-based language models show potential performance improvement. However, they face the challenge of designing the general-purpose framework for the language model, which will improve the performance of multi-task NLU and the generalized representation of knowledge. Combining these approaches may result in a more efficient and robust multi-task NLU. This SLR proposes building steps for a conceptual framework to achieve goals of enhancing the performance of language models in the field of NLU.

A Statistical Language Model for Pre-Trained Sequence Labeling: A Case Study on Vietnamese

By defining the computable word segmentation unit and studying its probability characteristics, we establish an unsupervised statistical language model (SLM) for a new pre-trained sequence labeling framework in this article. The proposed SLM is an optimization model, and its objective is to maximize the total binding force of all candidate word segmentation units in sentences under the condition of no annotated datasets and vocabularies. To solve SLM, we design a recursive divide-and-conquer dynamic programming algorithm. By integrating SLM with the popular sequence labeling models, Vietnamese word segmentation, part-of-speech tagging and named entity recognition experiments are performed. The experimental results show that our SLM can effectively promote the performance of sequence labeling tasks. Just using less than 10% of training data and without using a dictionary, the performance of our sequence labeling framework is better than the state-of-the-art Vietnamese word segmentation toolkit VnCoreNLP on the cross-dataset test. SLM has no hyper-parameter to be tuned, and it is completely unsupervised and applicable to any other analytic language. Thus, it has good domain adaptability.

Advancing smart building readiness: Automated metadata extraction using neural language processing methods

• We have demonstrated the suitability of a novel neural language processing method for automatically tagging building sensor metadata. • We achieve this by using a text pre-processor to tokenise building sensor names. • We train from scratch an unsupervised language model using sensor names from 152 commercial buildings. • We use the weights from this model to train a sensor type classifier. • We validate classification accuracy using data from 30 buildings. Digitalisation of the built environment provides multiple benefits such as operational and energy productivity improvements and supports the participation of buildings in the management of electricity networks. Automated methods to infer contextual information from building management systems and Internet of Things sensor metadata plays a significant role in this process. In this paper, we have studied the problem of transfer learning using text metadata to automatically tag building sensors with semantic tags. We demonstrate that state-of-the-art pre-trained neural language models are a promising approach which to the best of our knowledge have not been studied due to the lack of pre-processors to tokenise the texts. We develop a tokeniser based on the unigram language model capable of tokenising the idiosyncratic text found in building sensor metadata and use it to train from scratch a transformer based language model using sensor metadata from 152 buildings. The weights are then used to train a tagset classifier using transfer learning, and tested on 30 buildings. Metrics such as precision, recall and the Jaccard similarity coefficient have been used to evaluate the suitability of our results for various buildings. The proposed method can predict building tagsets with over 70% accuracy against a real world noisy dataset.

Unsupervised multi-sense language models for natural language processing tasks

Existing language models (LMs) represent each word with only a single representation, which is unsuitable for processing words with multiple meanings. This issue has often been compounded by the lack of availability of large-scale data annotated with word meanings. In this paper, we propose a sense-aware framework that can process multi-sense word information without relying on annotated data. In contrast to the existing multi-sense representation models, which handle information in a restricted context, our framework provides context representations encoded without ignoring word order information or long-term dependency. The proposed framework consists of a context representation stage to encode the variable-size context, a sense-labeling stage that involves unsupervised clustering to infer a probable sense for a word in each context, and a multi-sense LM (MSLM) learning stage to learn the multi-sense representations. Particularly for the evaluation of MSLMs with different vocabulary sizes, we propose a new metric, i.e., unigram-normalized perplexity (PPLu), which is also understood as the negated mutual information between a word and its context information. Additionally, there is a theoretical verification of PPLu on the change of vocabulary size. Also, we adopt a method of estimating the number of senses, which does not require further hyperparameter search for an LM performance. For the LMs in our framework, both unidirectional and bidirectional architectures based on long short-term memory (LSTM) and Transformers are adopted. We conduct comprehensive experiments on three language modeling datasets to perform quantitative and qualitative comparisons of various LMs. Our MSLM outperforms single-sense LMs (SSLMs) with the same network architecture and parameters. It also shows better performance on several downstream natural language processing tasks in the General Language Understanding Evaluation (GLUE) and SuperGLUE benchmarks.

Natural language predicts viral escape

Virology Viral mutations that evade neutralizing antibodies, an occurrence known as viral escape, can occur and may impede the development of vaccines. To predict which mutations may lead to viral escape, Hie et al. used a machine learning technique for natural language processing with two components: grammar (or syntax) and meaning (or semantics) (see the Perspective by Kim and Przytycka). Three different unsupervised language models were constructed for influenza A hemagglutinin, HIV-1 envelope glycoprotein, and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike glycoprotein. Semantic landscapes for these viruses predicted viral escape mutations that produce sequences that are syntactically and/or grammatically correct but effectively different in semantics and thus able to evade the immune system. Science , this issue p. [284][1]; see also p. [233][2] [1]: /lookup/doi/10.1126/science.abd7331 [2]: /lookup/doi/10.1126/science.abf6894

When Low Resource NLP Meets Unsupervised Language Model: Meta-Pretraining then Meta-Learning for Few-Shot Text Classification (Student Abstract)

Text classification tends to be difficult when data are deficient or when it is required to adapt to unseen classes. In such challenging scenarios, recent studies have often used meta-learning to simulate the few-shot task, thus negating implicit common linguistic features across tasks. This paper addresses such problems using meta-learning and unsupervised language models. Our approach is based on the insight that having a good generalization from a few examples relies on both a generic model initialization and an effective strategy for adapting this model to newly arising tasks. We show that our approach is not only simple but also produces a state-of-the-art performance on a well-studied sentiment classification dataset. It can thus be further suggested that pretraining could be a promising solution for few-shot learning of many other NLP tasks. The code and the dataset to replicate the experiments are made available at https://github.com/zxlzr/FewShotNLP.

GSAM: A deep neural network model for extracting computational representations of Chinese addresses fused with geospatial feature

Addresses are one of the most important geographical reference systems in natural languages. In China, due to the relatively backward address planning, there are a large number of non-standard addresses. This kind of unstructured text makes the management and application of Chinese addresses much more difficult. However, by extracting the computational representations of addresses, it can be structured and its related applications can be extended more conveniently. Therefore, this paper utilizes a deep neural language model from natural language processing (NLP) to automatically extract computational representations through an unsupervised address language model (ALM), which is trained in an unsupervised way and is suitable for a large-scale address corpus. We propose a solution to fuse addresses and geospatial features and construct a geospatial-semantic address model (GSAM) that supports a variety of downstream tasks. Our proposed GSAM constructing process consists of three phases. First, we build an ALM using bidirectional encoder representations from Transformers (BERT) to learn the addresses' semantic representations. Then, the fusion clustering results of the semantic and geospatial information are obtained by a high-dimensional clustering algorithm. Finally, we construct the GSAM based on the fused clustering results using novel fine-tuning techniques. Furthermore, we apply the extracted computational representation from GSAM to the address location prediction task. The experimental results indicate that the target task accuracy of the ALM is 90.79%, and the result of semantic geospatial fusion clustering strongly correlates with fine-grained urban neighbourhood area division. The GSAM can accurately identify clustering labels and the values of evaluation metrics are all above 0.96. We also demonstrate that our model outperforms purely ALM-based and word2vec-based models by address location prediction task.