ANTIPASTI: interpretable prediction of antibody binding affinity exploiting Normal Modes and Deep Learning DOI Creative Commons
Kevin Michalewicz, Mauricio Barahona, Barbara Bravi

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Dec. 23, 2023

Summary The high binding affinity of antibodies towards their cognate targets is key to eliciting effective immune responses, as well the use research and therapeutic tools. Here, we propose ANTIPASTI, a Convolutional Neural Network model that achieves state-of-the-art performance in prediction antibody using input representation antibody-antigen structures terms Normal Mode correlation maps derived from Elastic Models. This captures not only structural features but energetic patterns local global residue fluctuations. learnt representations are interpretable: they reveal similarities among targeting same antigen type, can be used quantify importance regions contributing affinity. Our results show imprint landscape, dominance cooperative effects long-range correlations between determine

Language: Английский

Navigating the landscape of enzyme design: from molecular simulations to machine learning DOI Creative Commons
Jiahui Zhou, Meilan Huang

Chemical Society Reviews, Journal Year: 2024, Volume and Issue: 53(16), P. 8202 - 8239

Published: Jan. 1, 2024

Global environmental issues and sustainable development call for new technologies fine chemical synthesis waste valorization. Biocatalysis has attracted great attention as the alternative to traditional organic synthesis. However, it is challenging navigate vast sequence space identify those proteins with admirable biocatalytic functions. The recent of deep-learning based structure prediction methods such AlphaFold2 reinforced by different computational simulations or multiscale calculations largely expanded 3D databases enabled structure-based design. While approaches shed light on site-specific enzyme engineering, they are not suitable large-scale screening potential biocatalysts. Effective utilization big data using machine learning techniques opens up a era accelerated predictions. Here, we review applications machine-learning guided We also provide our view challenges perspectives effectively employing design integrating molecular learning, importance database construction algorithm in attaining predictive ML models explore fitness landscape

Language: Английский

Citations

27

Decoding Catalysis by Terpene Synthases DOI Creative Commons
Joshua N. Whitehead, Nicole G. H. Leferink, Linus O. Johannissen

et al.

ACS Catalysis, Journal Year: 2023, Volume and Issue: 13(19), P. 12774 - 12802

Published: Sept. 15, 2023

The review by Christianson, published in 2017 on the twentieth anniversary of emergence field, summarizes foundational discoveries and key advances terpene synthase/cyclase (TS) biocatalysis (Christianson, D. W.

Language: Английский

Citations

44

Predicting Antimicrobial Peptides Using ESMFold-Predicted Structures and ESM-2-Based Amino Acid Features with Graph Deep Learning DOI

Greneter Cordoves‐Delgado,

César R. García‐Jacas

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: 64(10), P. 4310 - 4321

Published: May 13, 2024

Currently, antimicrobial resistance constitutes a serious threat to human health. Drugs based on peptides (AMPs) constitute one of the alternatives address it. Shallow and deep learning (DL)-based models have mainly been built from amino acid sequences predict AMPs. Recent advances in tertiary (3D) structure prediction opened new opportunities this field. In sense, graphs derived predicted peptide structures recently proposed. However, these are not correspondence with state-of-the-art approaches codify evolutionary information, and, addition, they memory- time-consuming because depend multiple sequence alignment. Herein, we presented framework create alignment-free graph representations generated ESMFold-predicted structures, whose nodes characterized acid-level information Evolutionary Scale Modeling (ESM-2) models. A attention network (GAT) was implemented assess usefulness AMP classification. To end, set comprised 67,058 used. It demonstrated that proposed methodology allowed build GAT generalization abilities consistently better than 20 non-DL-based DL-based The best were developed using 36- 33-layer ESM-2 Similarity studies showed best-built codified different chemical spaces, thus fused significantly improve general, results suggest esm-AxP-GDL is promissory tool develop good, structure-dependent, can be successfully applied screening large data sets. This should only useful classify AMPs but also for modeling other protein activities.

Language: Английский

Citations

17

Overview of AlphaFold2 and breakthroughs in overcoming its limitations DOI
Lei Wang,

Zehua Wen,

Shiwei Liu

et al.

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 176, P. 108620 - 108620

Published: May 15, 2024

Language: Английский

Citations

15

Leveraging protein structural information to improve variant effect prediction DOI Creative Commons
Lukas Gerasimavicius, Sarah A. Teichmann, Joseph A. Marsh

et al.

Current Opinion in Structural Biology, Journal Year: 2025, Volume and Issue: 92, P. 103023 - 103023

Published: Feb. 22, 2025

Despite massive sequencing efforts, understanding the difference between human pathogenic and benign variants remains a challenge. Computational variant effect predictors (VEPs) have emerged as essential tools for assessing impact of genetic variants, although their performance varies. Initially, sequence-based methods dominated field, but recent advances, particularly in protein structure prediction technologies like AlphaFold, led to an increased utilization structural information by VEPs aimed at scoring missense variants. This review highlights progress integrating into VEPs, showcasing novel models such AlphaMissense, PrimateAI-3D, CPT-1 that demonstrate improved evaluation. Structural data offers more interpretability, especially non-loss-of-function provides insights complex interactions vivo. As field utilizing biomolecular structures will be pivotal future VEP development, with breakthroughs protein-ligand protein-nucleic acid offering new avenues.

Language: Английский

Citations

2

Prediction of hemolytic peptides and their hemolytic concentration DOI Creative Commons
Anand Singh Rathore, Nishant Kumar, Shubham Choudhury

et al.

Communications Biology, Journal Year: 2025, Volume and Issue: 8(1)

Published: Feb. 4, 2025

Peptide-based drugs often fail in clinical trials due to their toxicity or hemolytic activity against red blood cells (RBCs). Existing methods predict peptides but not the concentration (HC50) required lyse 50% of RBCs. This study develops classification and regression models identify quantify activity. These train on 1926 with experimentally determined HC50 mammalian Analysis indicates that hydrophobic positively charged residues were associated higher Among models, including machine learning (ML), quantum ML, protein language a hybrid model combining random forest (RF) motif-based approach achieves highest area under receiver operating characteristic curve (AUROC) 0.921. Regression achieve Pearson correlation coefficient (R) 0.739 determination (R²) 0.543. outperform existing are implemented HemoPI2, web-based platform standalone software for designing desired values ( http://webs.iiitd.edu.in/raghava/hemopi2/ ).

Language: Английский

Citations

1

Identification and functional analysis of terpene synthases revealing the secrets of aroma formation in Chrysanthemum aromaticum DOI
Jian Zhong, Yu-Yuan Chen,

Huajin Shi

et al.

International Journal of Biological Macromolecules, Journal Year: 2024, Volume and Issue: 279, P. 135377 - 135377

Published: Sept. 6, 2024

Language: Английский

Citations

4

Prediction of electronic density of states in guanine-TiO2 adsorption model based on machine learning DOI Creative Commons
Yarkın A. Çetin, Benjamí Martorell, Francesc Serratosa

et al.

Deleted Journal, Journal Year: 2024, Volume and Issue: 1, P. 100008 - 100008

Published: June 2, 2024

The electronic density of states is a property the material that extensively used in quantum systems condensed matter physics. It refers to energy level electrons solid crystal. One most current ways compute it by Density Functional Tight Binding (DFTB), given geometry material. Nevertheless, this computation could be very computationally demanding, although applied some materials with reduced number atoms. This paper presents method deduce states, which based on neural network, thus, almost linear respect atoms Specifically, we have our metal oxide structure interacting nucleic base guanine. We focused stoichiometric and O-defective anatase TiO2 (101) surfaces. data set needed train network has been obtained DFTB+ numerical solver an initial molecular model, computed track time-dependent their associated states. validated predicted deduced DFTB tends similar, opening door other computations such introducing process generating analysis.

Language: Английский

Citations

3

Bioinformatics assisted construction of the link between biosynthetic gene clusters and secondary metabolites in fungi DOI
Huawei Lv,

Jia-Gui Tang,

Bin Wei

et al.

Biotechnology Advances, Journal Year: 2025, Volume and Issue: unknown, P. 108547 - 108547

Published: Feb. 1, 2025

Language: Английский

Citations

0

Quantum Mechanics Paradox in Protein Structure Prediction: Intrinsically Linked to Sequence yet Independent of it DOI Creative Commons
Sarfaraz K. Niazi

Deleted Journal, Journal Year: 2025, Volume and Issue: unknown, P. 100039 - 100039

Published: April 1, 2025

Language: Английский

Citations

0