About

Felipe Carvalho da Silva is an Assistant Professor and Extension Beef Specialist at North Carolina State University within the College of Agriculture and Life Sciences, Department of Animal Science. His long-term research goal is to improve reproductive efficiency in cattle operations by studying fetal-maternal interactions during early gestation, aiming to decrease early pregnancy loss, conserve natural resources, and enhance the economic viability of the US cattle industry. His current research focuses on the effects of dietary fatty acids and choline on embryonic elongation and maternal recognition of pregnancy. In addition to his research, he provides extension support to beef producers and extension personnel in North Carolina regarding reproductive management in beef operations. Dr. Silva holds a Ph.D. in Reproductive Biology from the University of Florida, an M.S. in Reproductive Physiology from North Dakota State University, and a B.Vet.Med in Veterinary Medicine from Sao Paulo State University – Jaboticabal. His areas of expertise include uterine biology, pregnancy, reproductive physiology, and beef cattle production.

Research topics

• Artificial Intelligence
• Computer Science
• Biology
• Biochemical engineering
• Genetics
• Physics
• Engineering
• Medicine
• Virology
• Computational biology
• Cell biology
• Data science

Selected publications

• ASPred: identification of antigen apecific B-cell receptors from single V(D)J sequences using large language models 4171

  The Journal of Immunology · 2025-11-01

  articleOpen access

  Abstract Description The rapid sequencing of antibody genes has accelerated vaccine development. However, predicting synthetic antibodies capable of binding and neutralizing novel antigens remains challenging due to a limited understanding of the rules of protein-protein interaction at the surface of an antigen to which its cognate antibody binds. While recent advances in single-cell sequencing of antibody-producing B-cells sequences have improved precision in mapping B-cell receptors to their cognate antigens, there remain additional challenges. We have developed a computational strategy, the Antibody Specificity Predictor (ASPred), with which we have trained two Large Language Models (LLMs) with known sequences of antigen-BCR pairs to predict antigen-specific BCRs from the total BCR repertoire. By leveraging pattern recognition capabilities of LLMs we successfully classify novel B-cell receptors with a challenge antigen not represented in the training set, without the need for preselecting the B cells by antigen binding. The properties of the top 10 predicted candidates were validated by coarse-grained molecular dynamics simulations. These results suggest that sufficient information exists in BCR-antigen sequence pairs for LLMs to reliably predict antigen-antibody interaction specificity, potentially opening new avenues for the computational design of synthetic antibodies for vaccine and therapeutic development. Funding Sources The work described in this report was supported by the National Institute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health (NIH) under the award number 1R01AI169543 Topic Categories Computational and Systems Immunology (COMP)

  PublisherOA PDFDOI

• ANABAG: Annotated Antibody Antigen dataset with unique features for Antibody Engineering Applications

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-07-07

  preprintOpen accessCorresponding

  Abstract The analysis and prediction of antibody–antigen (Ab–Ag) interactions often overlook critical structural features such as glycosylation, physical chemical conditions like pH and salt concentration, as well as the lack of standardized criteria for selecting complexes based on structural properties and sequence identity. Common practices in dataset construction rely on removing redundancy using sequence identity thresholds, which can inadvertently exclude complexes with alternative binding modes that share identical sequences. To enable more precise Ab–Ag modeling and antibody engineering, it is essential to incorporate richer structural and physical information into both physics-based and machine learning models. To address these limitations, we present ANABAG, a new curated dataset of Ab–Ag complexes annotated at the residue level with UniProt sequence information and enriched with a wide range of structural and physicochemical features. The dataset allows flexible filtering of complexes using a variety of descriptors available at both the complex and residue levels. Selected features are ready to use in machine learning workflows, while the structural files are compatible with antibody design and docking pipelines like Rosetta or Haddock. The complete dataset is available on Zenodo, and all accompanying scripts and usage documentation can be accessed via GitHub at https://github.com/DSIMB/anabag-handler.git .

  PublisherOA PDFDOI

• Fine-tuned protein language model identifies antigen-specific B cell receptors from immune repertoires

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-10-31

  preprintOpen access

  Scalable identification of antigen-specific antibodies from whole immune repertoire V(D)J sequences is a central challenge in biomedical engineering. We show that protein language models (PLMs) fine-tuned on antibody heavy-chain sequences can directly predict antigen specificity from unselected immune repertoires. We assessed our model, Antigen Specificity Predictor (ASPred), against SARS-CoV-2, influenza, and HIV-AIDS antigens, observing comparable predictive performance. In the whole immune repertoire V(D)J sequences of mice immunized with the SARS-CoV-2 spike protein's receptor-binding domain (RBD), ASPred identified antibody sequences specific to RBD. Several candidate sequences were validated, including one as a heavy chain-only nanobody with 20.7 nM dissociation constant. Molecular dynamics simulations supported the predicted interactions at coarse-grained and atomic levels. Benchmarking against Barcode-Enabled Antigen Mapping (BEAM) of B cell receptor sequence data had highly significant overlaps with ASPred predictions, suggesting scalability. The predicted SARS-CoV-2 binders differed substantially from training sequences, demonstrating generalization beyond sequence memorization. Together, we establish that heavy chain antibody sequences encode sufficient information for PLMs to infer specificity, offering a scalable framework for antibody discovery with broad applications.

  PublisherOA PDFDOI

• Biophysics of SARS-CoV-2 spike protein’s receptor-binding domain interaction with ACE2 and neutralizing antibodies: from computation to functional insights

  Biophysical Reviews · 2025-03-08 · 2 citations

  reviewOpen access1st authorCorresponding
  PublisherOA PDFDOI

• The role of charge regulation on casein–chitosan complexation at low pH

  Soft Matter · 2025-12-29 · 1 citations

  article

  ) contributions are attenuated by the same exponential Debye screening, the longer-range nature of charge regulation makes it the dominant effect in the investigated pH range, thereby ruling out ion-dipole interactions as the primary driving force. This work provided a quantitative and mechanistic confirmation that charge regulation was the dominant driving force for protein-polyelectrolyte association on the "wrong side" of the isoelectric point, offering fundamental insights for the rational design of biomolecular complexes.

  PublisherDOI

• ANABAG: Annotated Antibody–Antigen Data Set with Unique Features for Antibody Engineering Applications

  Journal of Chemical Information and Modeling · 2025-10-17 · 1 citations

  article

  The analysis and prediction of antibody-antigen (Ab-Ag) interactions often overlook critical structural features such as glycosylation and important physicochemical conditions like pH and salt concentration. Additionally, the field lacks standardized criteria for selecting complexes based on structural properties and sequence identity. Common practices in data set construction rely on removing redundancy using sequence identity thresholds, which can inadvertently exclude complexes with alternative binding modes that share identical sequences. To enable more precise Ab-Ag modeling and antibody engineering, it is essential to incorporate richer structural and physical information into both physics-based and machine learning models. To address these limitations, we present ANABAG, a new curated data set of Ab-Ag complexes annotated at the residue level with UniProt sequence information and enriched with a wide range of structural and physicochemical features. The data set allows flexible filtering of complexes using a variety of descriptors available at both the complex and residue levels. Selected features are ready to use in machine learning workflows, while the structural files are compatible with antibody design and docking pipelines like Rosetta or Haddock. The complete data set is available on Zenodo at https://zenodo.org/records/17065788, and all accompanying scripts and usage documentation can be accessed via GitHub at https://github.com/DSIMB/anabag-handler.git.

  PublisherDOI

• Studying self-assembly of norovirus capsid by a combination of <i>in silico</i> methods

  bioRxiv (Cold Spring Harbor Laboratory) · 2024-01-23

  preprintOpen access

  Abstract Understanding how macromolecular assembly occurs is a fundamental and challenging problem because spontaneous, precise assembly is at the center of most biological processes. It is an elaborate process that requires non-covalent stable interactions between partners to stabilize the desired architecture for a specific purpose. One of the advantages of virus models is that under adequate conditions capsid proteins can be efficiently assembled in vitro in the absence of any other component, providing simplified experimental models that can be rigorously characterized. The present study aims at describing the initial steps of molecular self-assembly of norovirus-like particles (NoVLPs, composed solely of the major norovirus capsid protein VP1), by combining in silico computational approaches to explore complementary physical properties. We show that this strategy allows not only recapitulating but also revising a former NoVLP assembly model. Our approach can be applied and extended to other problems in macromolecular assembly.

  PublisherOA PDFDOI

• Constant-pH Simulation Methods for Biomolecular Systems

  Elsevier eBooks · 2023-06-01 · 2 citations

  book-chapter1st authorCorresponding
  PublisherDOI

• NS1 from Two Zika Virus Strains Differently Interact with a Membrane: Insights to Understand Their Differential Virulence

  Journal of Chemical Information and Modeling · 2023-02-13 · 7 citations

  article

  Zika virus (ZIKV) from Uganda (UG) expresses a phenotype related to fetal loss, whereas the variant from Brazil (BR) induces microcephaly in neonates. The differential virulence has a direct relation to biomolecular mechanisms that make one strain more aggressive than the other. The nonstructural protein 1 (NS1) is a key viral toxin to comprehend these viral discrepancies because of its versatility in many processes of the virus life cycle. Here, we aim to examine through coarse-grained models and molecular dynamics simulations the protein–membrane interactions for both NS1ZIKV-UG and NS1ZIKV-BR dimers. A first evaluation allowed us to establish that the NS1 proteins, in the membrane presence, explore new conformational spaces when compared to systems simulated without a lipid bilayer. These events derive from both differential coupling patterns and discrepant binding affinities to the membrane. The N-terminal domain, intertwined loop, and greasy finger proposed previously as binding membrane regions were also computationally confirmed by us. The anchoring sites have aromatic and ionizable residues that manage the assembly of NS1 toward the membrane, especially for the Ugandan variant. Furthermore, in the presence of the membrane, the difference in the dynamic cross-correlation of residues between the two strains is particularly pronounced in the putative epitope regions. This suggests that the protein–membrane interaction induces changes in the distal part related to putative epitopes. Taken together, these results open up new strategies for the treatment of flaviviruses that would specifically target these dynamic differences.

  PublisherDOI

• Differences between Omicron SARS-CoV-2 RBD and other variants in their ability to interact with cell receptors and monoclonal antibodies

  bioRxiv (Cold Spring Harbor Laboratory) · 2022-01-31 · 6 citations

  preprintOpen accessSenior authorCorresponding

  Abstract SARS-CoV-2 has caused immeasurable damage worldwide and available treatments with high efficacy are still scarce. With the continuous emergence of new variants of the virus, such as Omicron, Alpha, Beta, Gamma, and Delta - the so-called variants of concern, the available therapeutic and prevention strategies had to return to the experimental trial to verify their effectiveness against them. This work aims to expand the knowledge about the SARS-CoV-2 receptor-binding domain (RBD) interactions with cell receptors and monoclonal antibodies (mAbs). Special attention is given to the Omicron variant and its comparison with the others, including its sublineage BA.2 and two new ones (B.1.640.1 and B.1.640.2/IHU) recently found in France. By using constant-pH Monte Carlo simulations, the free energy of interactions between the SARS-CoV-2 receptor-binding domain (RBD) from different variants and several partners (Angiotensin-Converting Enzyme-2 (ACE2) polymorphisms and several mAbs) were calculated. It was evaluated both the impact of mutations for the RBD-ACE2 and how strongly each of mAb can bind to the virus RBD, which can indicate their therapeutic potential for neutralization. RBD-ACE2-binding affinities were higher for two ACE2 polymorphisms typically found in Europeans (rs142984500 and rs4646116), indicating that these types of polymorphisms may be related to genetic susceptibility to COVID-19. The antibody landscape was computationally investigated with the largest set of mAbs so far in the literature. From the 33 studied binders, groups of mAbs were identified with weak (e.g. S110 and Ab3b4), medium (e.g. CR3022), and strong binding affinities (e.g. P01’’’, S2K146 and S230). All the mAbs with strong binding capacity could also bind to the RBD from SARS-CoV-1, SARS-CoV-2 wt, and all studied variants. These mAbs and especially their combination are amenable to experimentation and clinical trials because of their high binding affinities and neutralization potential for current known virus mutations and a universal coronavirus.

  PublisherOA PDFDOI

Frequent coauthors

• Samuela Pasquali

  Université Paris Cité

  120 shared

• Philippe Derreumaux

  Institut Universitaire de France

  120 shared

• Aatto Laaksonen

  113 shared

• Catherine Etchebest

  Inserm

  86 shared

• Carolina Corrêa Giron

  Universidade de São Paulo

  60 shared

• Sergio Alejandro Poveda Cuevas

  Goethe University Frankfurt

  46 shared

• Fabio Sterpone

  Centre National de la Recherche Scientifique

  35 shared

• Bo Jönsson

  Lund University

  13 shared

Education

• Docent (Livre-docente), Biomolecular Sciences - School of Pharmaceutical Sciences at Ribeirão Preto

  Universidade de São Paulo

  2008

• Ph.D. in Theoretical Chemistry, Theoretical Chemistry

  Lunds Universitet

  2000