HLA class I binding prediction via convolutional neural networks

Vang, Yeeleng Scott; Xie, Xiaohui

doi:10.1101/099358

Cited by 18 publications

(34 citation statements)

References 44 publications

(37 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They can also embedded residues, proteins and chemicals into vectors as the input for model training without artificial design of features or expert knowledge. It had been used to predict HLA binding proteins, antimicrobial peptides and drug targets [36][37][38][39]. Our current research validates its capacity in PSP prediction.…”

Section: Discussionsupporting

confidence: 66%

Prediction of liquid-liquid phase separation proteins using machine learning

Sun

et al. 2019

Preprint

View full text Add to dashboard Cite

The liquid-liquid phase separation (LLPS) of bio-molecules in cell underpins the formation of membraneless organelles, which are the condensates of protein, nucleic acid, or both, and play critical roles in cellular functions. The dysregulation of LLPS might be implicated in a number of diseases. Although the LLPS of biomolecules has been investigated intensively in recent years, the knowledge of the prevalence and distribution of phase separation proteins (PSPs) is still lag behind. Development of computational methods to predict PSPs is therefore of great importance for comprehensive understanding of the biological function of LLPS. Here, a sequencebased prediction tool using machine learning for LLPS proteins (PSPredictor) was developed. Our model can achieve a maximum 10-CV accuracy of 96.03%, and performs much better in identifying new PSPs than reported PSP prediction tools. As far as we know, this is the first attempt to make a direct and more general prediction on LLPS proteins only based on sequence information.

show abstract

Section: Discussionsupporting

confidence: 66%

Prediction of liquid-liquid phase separation proteins using machine learning

Sun

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Kim and Han [11] proposed a DCNN-based pan specific model and showed the power of DCNN in revealing the locally-clustered patterns in peptide-HLA complexes. Vang and Xie [15] developed a DCNN based allele-specific model with a novel distributed representation of amino acids and has validated the importance of the CNN part in improving the performance. In our previous work, we developed DeepSeqPan [10], which is also a pan-specific model based on two separate DCNN encoders to extract high-level features of both the peptide and the HLA sequence.…”

Section: Introductionmentioning

confidence: 99%

“…There are two major categories of models for MHC-peptide binding affinity prediction: pan-specific models [2,4,5,[7][8][9][10][11] such as NetMHCpan [4,5,7,9], DeepSeqPan [10], Pickpocket [8], and ACME [2] and allele-specific models [1,[12][13][14][15] such as ANN [13,14], ARB [16], SMM [12]. Panspecific models are usually trained on datasets of multiple alleles while allele-specific models are developed for specific alleles.…”

Section: Introductionmentioning

confidence: 99%

Attention mechanism-based deep learning pan-specific model for interpretable MHC-I peptide binding prediction

Jin

Nasiri

et al. 2019

Preprint

View full text Add to dashboard Cite

Accurate prediction of peptide binding affinity to the major histocompatibility complex (MHC) proteins has the potential to design better therapeutic vaccines. Previous work has shown that pan-specific prediction algorithms such as NetMHCPan, DeepSeqPan and ACME can achieve better prediction performance than other approaches. Here, we propose DeepAttentionPan, an improved pan-specific model, based on convolution neural networks and an attention mechanism for more flexible, stable and interpretable MHC-I binding prediction. Our model can deal with peptides of different lengths with improved prediction performance due to the attention modules Our ensemble model consisting of 20 trained networks from two rounds of 10-fold cross-validation brings higher and more stabilized prediction performance. Extensive tests on the IEDB benchmark data show that our method outperforms other state-of-the-art prediction algorithms on 21 test allele datasets. Visualization analysis of the attention weights of our model further demonstrates the capability of our attention module to capture critical binding positions of the peptides, which leads to mechanistic understanding of MHC-peptide binding with high alignment with experimentally verified results. Furthermore, we have also shown that with transfer learning, our pan model can be fine-tuned for alleles with few samples to achieve additional performance improvement. DeepAttentionPan is freely available as an open source software at https://github.com/jjin49/DeepAttentionPan. Author summary:Human leukocyte antigen (HLA) proteins are classes of proteins that are responsible for immune system regulation in humans. The peptides are short chains of amino acids. HLA class I group present peptides from inside the cell to the cell surface for scrutiny by T cell receptors. For instance, if the cell is infected by a virus, the HLA system will bind to the peptides derived from viral proteins and bring them to the surface of the cell so that the cell can be destroyed by the immune system. Since the HLA genes exhibit extensive polymorphism, there are many HLA alleles binding to different peptides. And this diversity represents challenges in predicting binders for different HLA alleles, which are important in vaccine designs and characterization of immune responses. Before computational algorithms are used to predict the binding relationships of HLApeptide pairs, scientists need to conduct costly biological experiments to do preliminary screening among a number of peptides and need to use mutant experiments to identify key peptide positions that contribute to the binding. While previous computational methods have been proposed to predict the binding affinity, identifying the binding anchors is not well addressed. Here we developed a deep neural network models with the attention mechanism to learn the binding relationships automatically in an end-to-end way. Our models are able to identify the important binding positions of the peptide sequence by learning the positional importance distribution, w...

show abstract

“…Letter-based representations are ubiquitous in addressing complicated functions owing 55 to their simplicity, applicability, and accuracy in finding aligned domains in a sequence 56 [14][15][16][17] or within a larger structure [18][19][20]. Several Machine Learning (ML) models to 57 predict functionality using deep-learning, NNs, feature representation, and pattern 58 analyses such as DeepMHC and NetMHCpan among others [21][22][23], have been 59 developed by using the data in the Immuno-Epitope Database (IEDB) Analysis resource 60…”

mentioning

confidence: 99%

VSEPRnet: Physical structure encoding of sequence-based biomolecules for functionality prediction: Case study with peptides

Rath

Francis-Landau

et al. 2019

Preprint

View full text Add to dashboard Cite

1Predicting structure-dependent functionalities of biomolecules is crucial for accelerating 2 a wide variety of applications in drug-screening, biosensing, disease-diagnosis, and 3 therapy. Although the commonly used structural "fingerprints" work for biomolecules in 4 traditional informatics implementations, they remain impractical in a wide range of 5 machine learning approaches where the model is restricted to make data-driven 6 decisions. Although peptides, proteins, and oligonucleotides have sequence-related 7 propensities, representing them as sequences of letters, e.g., in bioinformatics studies, 8 causes a loss of most of their structure-related functionalities. Biomolecules lacking 9 sequence, such as polysaccharides, lipids, and their peptide conjugates, cannot be 10 screened with models using the letter-based fingerprints. Here we introduce a new 11 fingerprint derived from valence shell electron pair repulsion structures for small peptides 12 that enables construction of structural feature-maps for a given biomolecule, regardless 13 of the sequence or conformation. The feature-map introduced here uses a simple 14 encoding derived from the molecular graph -atoms, bonds, distances, bond angles, etc., 15 that make up each of the amino acids in the sequence, allowing a Residual Neural 16 network model to take greater advantage of information in molecular structure. We make 17 use of the short peptides binding to Major-Histocompatibility-Class-I protein alleles that 18 are encoded in terms of their extended structures to predict allele-specific binding-19 affinities of test-peptides. Predictions are consistent, without appreciable loss in accuracy 20 between models for different length sequences, marking an improvement over the current 21 models. Biological processes are heterogeneous interactions, which justifies encoding all 22 biomolecules universally in terms of structures and relating them to their functionality. The 23 capabilities facilitated by the model expands the paradigm in establishing structure-24 function correlations among small molecules, short and longer sequences including large 25 biomolecules, and genetic conjugates that may include polypeptides, polynucleotides, 26RNAs, lipids, peptidoglycans, peptido-lipids, and other biomolecules that could be 27 implemented in a wide range of medical and nanobiotechnological applications in the 28 future. 29

show abstract

HLA class I binding prediction via convolutional neural networks

Cited by 18 publications

References 44 publications

Prediction of liquid-liquid phase separation proteins using machine learning

Prediction of liquid-liquid phase separation proteins using machine learning

Attention mechanism-based deep learning pan-specific model for interpretable MHC-I peptide binding prediction

VSEPRnet: Physical structure encoding of sequence-based biomolecules for functionality prediction: Case study with peptides

Contact Info

Product

Resources

About