xTrimoABFold: De novo Antibody Structure Prediction without MSA

Gong, Xumeng; Li, Shaochuan; Yang, Bing; Sun, YiWu; Shi, Chuan; Li, Hui; Wang, Yangang; Yang, Cheng; Song, Lixin

doi:10.48550/arxiv.2212.00735

Cited by 3 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, there is no available rigid structure in the actual scenario, therefore, we applied the current state-of-the-art protein folding prediction model to generate the rigid structure. We employ xTrimoABFold (Wang et al, 2022) to predict structures for antibody heavy chain and light chain, and employ AlphaFold2 (Jumper et al, 2021) to predict a structure for the antigen chain if there are no available structures experimentally measured by X-ray crystallography, cryo-EM, NMR spectroscopy, dual polarization interferometry, or other methods.…”

Section: Methodsmentioning

confidence: 99%

“…The stat-of-the-art computational protein folding models provide alternative solutions and provide protein structures from single-chain primary sequences with atomic accuracy. In this paper, we predict the structures of antibody’s light chains and heavy chains by xTrimoABFold (Wang et al, 2022) and antigen’s structures by AlphaFold2 (Jumper et al, 2021). To eliminate the effects of MSAs, we use the same MSAs for antigen sequences on AlphaFold2 and AlphaFold-Multimer.…”

Section: Methodsmentioning

confidence: 99%

“…Other structure prediction methods, such as OmegaFold (Wu et al, 2022), Helixfold-single (Fang et al, 2022), ESM-Fold (Lin et al, 2022), employ large-scale protein language models to replace computationally extensive MSA searching and achieve comparable performance with AlphaFold2. Models designed specifically for antibody structure prediction, such as DeepAb (Ruffolo et al, 2022), ABlooper (Abanades et al, 2022a), IgFold (Ruffolo and Gray, 2022), xTrimoABFold (Wang et al, 2022), ImmuneBuilder (Abanades et al, 2022b) have recently been developed. In particular, xTrimoABFold uses a pretrained antibody language model and crossmodal template searching algorithm to obtain the most accurate heavy or light chain structure of an antibody.…”

Section: Backgrounds and Related Workmentioning

confidence: 99%

See 2 more Smart Citations

xTrimoDock: Rigid Protein Docking via Cross-Modal Representation Learning and Spectral Algorithm

Luo¹,

Li²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

Protein-protein interactions are the basis for the formation of protein complexes which are essential for almost all cellular processes. Knowledge of the structures of protein complexes is of major importance for understanding the biological function of these protein-protein interactions and designing protein drugs. Here we address the problem of rigid protein docking which assumes no deformation of the involved proteins during interactions. We develop a method called, xTrimoDock, which leverages a cross-modal representation learning to predict the protein distance map, and then uses a spectral initialization and gradient descent to obtain the roto-translation transformation for docking. We show that, on antibody heavy-chain and light-chain docking, and antibody-antigen docking, xTrimoDock consistently outperforms the state-of-the-art such as AlphaFold-Multimer and HDock, and can lead to as much as a 10% improvement in DockQ metric. xTrimoDock has been applied as a useful tool in protein drug design at BioMap.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Backgrounds and Related Workmentioning

confidence: 99%

See 1 more Smart Citation

xTrimoDock: Rigid Protein Docking via Cross-Modal Representation Learning and Spectral Algorithm

Luo¹,

Li²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…However, a more recent study of pLM pretraining strategies suggests that Masked Language Modeling (MLM) is particularly effective for structure-based modeling, injecting many structurally correlated patterns into the pLM latent space (22). Whereas this gives insight into the success of protein folding models (21,(23)(24)(25)(26)(27)(28)(29)), we assume that the optimal latent space for annotation should more closely correlate with more abstract concepts like "protein function" and "biological process." We also surmise that generating proteins for specific properties can be actualized by a closer relationship between a "property" latent space and sequence latent space.…”

Section: Introductionmentioning

confidence: 99%

Annotation Vocabulary (Might Be) All You Need

Hallee,

Rafailidis,

Horger

et al. 2024

Preprint

View full text Add to dashboard Cite

Protein Language Models (pLMs) have revolutionized the computational modeling of protein systems, building numerical embeddings that are centered around structural features. To enhance the breadth of biochemically relevant properties available in protein embeddings, we engineered the Annotation Vocabulary, a transformer readable language of protein properties defined by structured ontologies. We trained Annotation Transformers (AT) from the ground up to recover masked protein property inputs without reference to amino acid sequences, building a new numerical feature space on protein descriptions alone. We leverage AT representations in various model architectures, for both protein representation and generation. To showcase the merit of Annotation Vocabulary integration, we performed 515 diverse downstream experiments. Using a novel loss function and only $3 in commercial compute, our premier representation model CAMP produces state-of-the-art embeddings for five out of 15 common datasets with competitive performance on the rest; highlighting the computational efficiency of latent space curation with Annotation Vocabulary. To standardize the comparison of de novo generated protein sequences, we suggest a new sequence alignment-based score that is more flexible and biologically relevant than traditional language modeling metrics. Our generative model, GSM, produces high alignment scores from annotation-only prompts with a BERT-like generation scheme. Of particular note, many GSM hallucinations return statistically significant BLAST hits, where enrichment analysis shows properties matching the annotation prompt - even when the ground truth has low sequence identity to the entire training set. Overall, the Annotation Vocabulary toolbox presents a promising pathway to replace traditional tokens with members of ontologies and knowledge graphs, enhancing transformer models in specific domains. The concise, accurate, and efficient descriptions of proteins by the Annotation Vocabulary offers a novel way to build numerical representations of proteins for protein annotation and design.

show abstract