Computational methods are required to study peptides and gain insights into rational design strategies. Open-source frameworks can accelerate the analysis of complex peptides, reducing human error. Some open-source protocols for peptide research will be presented. PepFuNN is a Novo Nordisk package for running cheminformatics analysis of peptides. BILN is a peptide-focused line notation, and pyPept is a Python package for representing natural and modified peptides using BILN, facilitating conversion to formats such as FASTA and HELM. Finally, mPARCE is an evolutionary protocol to design peptide sequences using Rosetta and external scoring functions. The main focus is the inclusion of non-canonical amino acids on previously generated models or experimentally solved structures.

Background: Peptide self-assembly is a key phenomenon in molecular science, underpinning advancements in regenerative medicine, biosensing, and nanotechnology. While natural peptides offer biocompatibility, they are limited by low structural diversity, biological instability, and inconsistent self-assembly behavior. Peptides composed of non-natural amino acids, such as fluorinated β-diarylamino acids, offer a promising route to overcome these limitations and engineer next-generation biomaterials with enhanced functional properties.

Objective: To investigate how stereochemistry and environmental factors influence the conformational dynamics and aggregation behavior of non-natural tetrapeptides containing fluorinated β-diarylamino acids and L-alanine.

Methods: We performed classical molecular dynamics (cMD) and accelerated molecular dynamics (aMD) simulations using the Amber software suite with the ff19SB force field in explicit TIP3P water. Four stereo chemically distinct tetrapeptides were studied at concentrations of 681 μM (9 copies) and 1860 μM (30 copies) over 2.8 μs (cMD) and 0.3 μs (aMD) of cumulative simulation time. Structural analyses focused on secondary structure motifs, compactness, flexibility, hydrogen bonding networks, and solvent-accessible surface area (SASA).

Results: Aggregation propensity was highly dependent on peptide stereochemistry and concentration. Turn motifs were predominant across all peptides, with variations in hydrogen bonding and structural compactness. aMD simulations revealed deeper aggregation states with up to 27.5 intermolecular hydrogen bonds and significantly reduced SASA values. D1Cl1 consistently showed strong aggregation due to a rigid β-turn geometry and stable π–π interactions, while D2Cl2 displayed prominent clustering under aMD, linked to its stereochemical configuration.

Conclusion: Our findings demonstrate that stereochemical design and simulation strategy critically influence peptide self-assembly. The integration of non-natural amino acids enables the formation of stable supramolecular aggregates with desirable properties, providing a rational framework for the design of advanced peptide-based biomaterials for therapeutic and technological applications.

It is highly desirable to predict the developability of therapeutic antibodies via in silico analysis, both to speed up the candidate selection process and to avoid issues such as aggregation and fast clearance during clinical development. Here we explore the use of molecular descriptors extracted from molecular dynamics simulations, in combination with machine learning algorithms, to predict the outcomes of several developability assays. This work demonstrates that antibody biophysical properties are driven by a complex combination of molecular features, but that useful predictions can be made using an entirely in silico approach.

Predicting potential liabilities, aggregation, viscosity etc. is of importance in monoclonal antibody development. Computational property prediction methods are routinely used in the selection and optimization of candidate antibodies. High quality property prediction involves prediction of ensembles of 3D structures at specified pH to reduce sensitivity to single conformational states.

We present 3dpredict/Ab which calculates ensemble-based predictions of antibody developability descriptors and putative liabilities. 3dpredict/Ab allows for out-of-the-box SaaS automation and integration of such complex simulations of hundreds or thousands of sequences.

Knowledge of the precise coordinates of antibody epitopes are crucial in vaccine design, therapeutic antibody engineering, and optimizing diagnostic and therapeutic applications in molecular medicine. Experimental approaches to capture antibody epitope are time consuming, laborious, and expensive. Therefore, not amenable to be applied on a large-scale for comprehensive BCE mapping and screening.

Herein, we use ML based epitope profiling of SARS-Cov-2 Spike variants to identify conserved and variant specific epitopes. We then applied a mutation policy to generate an artificial spike protein bearing a set of epitope with broad coverage of variants. Finally, we generated mRNA-LNP vaccines encoding for the AI designed antigen and compared its neutralization effect on different SARS variant after immunization in the mouse. The newly generated mRNA-LNP vaccines show broader coverage with respect to the available SARS-Cov-2 vaccines.

Antibodies are key immunotherapeutics that recognize antigens via paratope–epitope interactions. Sequence–diverse paratopes can bind to the same epitope with similar affinity, but the structural principles enabling such convergence remain unclear, hindered by limited high–resolution structures and computational models that fail to capture antibody–antigen dynamics. We resolved the binding sites of five CDRH3–diverse, affinity–verified Trastuzumab variants using cryo–EM and HDX–MS, and compared them with computational models before and after molecular dynamics simulations. Analysis of MD trajectories identified structural parameters that distinguish antibody variants according to experimentally measured binding strength, revealing features invisible in static structures alone. These results provide new insights into how paratope diversity translates into binding specificity and represent a critical step toward defining structural rules of antibody recognition.

VHH nanobody discovery traditionally relies on animal immunization or large-scale library screening—methods that are slow, expensive, and often ineffective for challenging targets such as GPCRs. We present a fully de novo computational pipeline for epitope-directed VHH design that integrates generative backbone modeling (RFdiffusion), deep learning–based sequence optimization (ByProt), and AlphaFold2-guided interface refinement within an iterative design–build–test–learn framework. Using LGR5 as a model target, we progressed from in silico design to functional binders without structural templates or existing binders. Across three design cycles, millions of candidates were computationally filtered to yield epitope- specific nanobodies with nanomolar affinity and high thermal stability (Tm > 65°C). Cryo-EM structural determination confirmed atomic-level agreement between designed and experimentally determined structures (RMSD ≈ 2 Å). This work establishes a generalizable paradigm for rapid, on-demand generation of therapeutic antibodies against membrane proteins and other challenging targets.