Library of peptides: Analysis of a complex reactive mixture with the use of statistical physics and inference
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "Library of peptides".
Mira: Here's the gist of the paper. Self-assembled mixtures enable the exploration of vast chemical diversity, yet characterizing the hundreds of coexisting species they contain remains a major challenge Here, we introduce a framework that bridges statistical physics and analytical chemistry to deconvolute complex self-assembled mixtures from partial spectral data Framework Introduction The proposed method expresses the chemical potential of each compound as a function of a reduced set of parameters—intrinsic building-block energies and pairwise coupling terms between neighboring building blocks in the sequence This model is coupled with the law of mass action to yield equilibrium concentrations across the entire reactive network The model parameters are then inferred by fitting predicted spectroscopic signals to experimentally measured partial data.
Kai: Modeling Self-Assembly Energetics The approach is inspired by Potts models of protein folding, where the energy of each conformer is expressed as a function of the intrinsic energies of the amino acids and coupling terms between neighbors, encoding spatial proximity The concentrations are then expressed as a function of their relative energy, the reactive network topology, and initial concentrations of building blocks.
Mira: Experimental Setup and Data The study uses dynamic combinatorial libraries (DCLs) of peptides as a model system Experiments involve a scaffold with four reactive sites and a set of two building blocks, such as alanine and leucine The relative composition is monitored by RP-HPLC at equilibrium, where peaks correspond to groups of peptides.
Kai: Inference and Validation The method requires determining the set of parameters X from a reduced set of experiments to maximize the likelihood with available partial data The distance metric used is the sum over all peaks obtained in the experimental subset of the squared logarithm of the ratio between peak area obtained from X and from experiments.
Paper summary: Kai: So, to recap what we've heard about this paper, it’s about a challenge in characterizing self-assembled mixtures where standard techniques give us overlapping signals that conventional deconvolution methods struggle to resolve because they ignore the concentration correlations in the reactive network.
Mira: They propose a framework that uses statistical physics models, specifically inspired by Potts models of protein folding, to express the chemical potential of each compound based on intrinsic building-block energies and coupling terms between neighbors in the sequence.
Lev: This model is then coupled with the law of mass action to calculate equilibrium concentrations across the entire reactive network, and those model parameters are inferred by fitting predicted spectroscopic signals against experimentally measured partial data.
Kai: The system they use is a dynamic combinatorial library of peptides as a model, and they show that this approach allows them to determine the full composition of this DCL, which has eighty-one species, from just thirteen overlapping RP-HPLC peaks at equilibrium.
Mira: They emphasize that their method takes into account the combinatorial nature of the ensemble of chemical species and the concentration correlations imposed by the reactive network, assuming those network relationships are known to them.
Lev: The validation involved dividing their experimental data into a fitting set and a validation set, showing that increasing model complexity by adding terms like transverse coupling, lateral coupling, and cross-coupling progressively reduces the distance metric used for comparison.
Kai: Their key result is that in their studied DCL of eighty-one peptides, the lateral couplings dominate the assembly energetics, which is important context for understanding how these systems organize themselves spatially <ref:2611.17130#pg3>.
Conclusion: Mira: Looking at this paper, "Library of peptides: Analysis of a complex reactive mixture with the use of statistical physics and inference," it provides a concrete strategy for deconvolving complex self-assembled mixtures from partial spectral data using statistical physics.
Kai: The authors argue that this bridge between physical models and analytical chemistry allows them to reconstruct the full composition of a DCL with eighty-one species from just thirteen HPLC peaks, which is a lot of information recovered.
Lev: For someone who builds real hardware, the implication is that if you can reduce the required parameter space by a factor of three hundred that means you have a much more tractable inference problem on real systems.
Mira: It suggests that designing self-assembly protocols can become less reliant on exhaustive screening because this method enables the determination of full compositions from much sparser experimental data points.
Kai: The title itself highlights how they are using statistical physics to analyze these mixtures, moving beyond simple fitting to incorporate the underlying physical constraints imposed by the reactive network.
Lev: What's interesting is that they explicitly state their limitation: their method relies on knowing the underlying network topology beforehand, which is a strong assumption for any real-world system where you don't know everything about it a priori.
Mira: So, while it’s powerful for these specific combinatorial systems, its broad application depends heavily on whether we can accurately map out that reactive network structure first.
Laura Messio, Taleen Peker, Roba Moumné, ğ Helene Berthoumieux
Sorbonne University · CNRS Laboratoire de Physique Thພorique de la Matiere Condensພe
physics.chem-ph, physics.data-an
Submitted: 2026-11-12
Updated: 2026-11-12
Comments: 15 pages, 6 figures
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 72/100
The gist: Self-assembled mixtures enable the exploration of vast chemical diversity, yet characterizing the hundreds of coexisting species they contain remains a major challenge Here…
Key concepts
- Chemical Potential
- This concept describes the energy change associated with adding one more molecule of a specific chemical species to a system. In this model, it is expressed as a function of intrinsic building-block energies and how neighboring molecules interact spatially within the self-assembled structure.
- Potts Models
- Inspired by protein folding models, these statistical physics models are used to describe the energy landscape of self-assembly. They treat each molecule (conformer) as having an energy based on its neighbors' positions and intrinsic properties, helping to predict how molecules will arrange themselves.
- Dynamic Combinatorial Libraries (DCLs)
- These are large collections of chemically possible molecules, like peptides, that can be synthesized. The study used a DCL of 81 peptides as a test system to show how the proposed framework can efficiently determine the mixture composition from experimental results.
- Inference and Distance Metric
- This is the process of using limited experimental data to find the best possible set of model parameters. A distance metric calculates how well a predicted spectrum matches real data; minimizing this distance ensures that the chosen parameters accurately reflect the actual mixture composition.
Terminology
Summary
Self-assembled mixtures enable the exploration of vast chemical diversity, yet characterizing the hundreds of coexisting species they contain remains a major challenge Here, we introduce a framework that bridges statistical physics and analytical chemistry to deconvolute complex self-assembled mixtures from partial spectral data
Framework Introduction
The proposed method expresses the chemical potential of each compound as a function of a reduced set of parameters—intrinsic building-block energies and pairwise coupling terms between neighboring building blocks in the sequence This model is coupled with the law of mass action to yield equilibrium concentrations across the entire reactive network The model parameters are then inferred by fitting predicted spectroscopic signals to experimentally measured partial data.
Modeling Self-Assembly Energetics
The approach is inspired by Potts models of protein folding, where the energy of each conformer is expressed as a function of the intrinsic energies of the amino acids and coupling terms between neighbors, encoding spatial proximity The concentrations are then expressed as a function of their relative energy, the reactive network topology, and initial concentrations of building blocks.
Experimental Setup and Data
The study uses dynamic combinatorial libraries (DCLs) of peptides as a model system Experiments involve a scaffold with four reactive sites and a set of two building blocks, such as alanine and leucine The relative composition is monitored by RP-HPLC at equilibrium, where peaks correspond to groups of peptides.
Inference and Validation
The method requires determining the set of parameters X from a reduced set of experiments to maximize the likelihood with available partial data The distance metric used is the sum over all peaks obtained in the experimental subset of the squared logarithm of the ratio between peak area obtained from X and from experiments. Validation confirms that increasing model complexity, by including proper energies, transverse coupling terms, lateral coupling terms, and cross-coupling terms, progressively decreases the distance metric. p. 8 To evaluate the accuracy of the model, we progressively enlarge our set of optimized parameters:
first, we only consider the proper energies Ei (X = E) and compute and mimize the distance ∆χ for
the training experiments following the steps described in the previous section. Then, we repeat the
optimization when including the transverse coupling terms, Tij , occurring between sites (1, 8), and (3,
6), as shown in Fig.Where the paper says it
Key Findings
The analysis demonstrates that lateral couplings dominate the assembly energetics in the studied DCL of 81 peptides The resulting model requires a significantly reduced number of parameters compared to the total number of possible molecules, achieving a reduction by a factor of approximately 300 for 20 building blocks. This efficiency allows for the full resolution of the mixture composition from just 13 overlapping peaks. p. 0 We demonstrate
this approach using dynamic combinatorial libraries (DCLs) of peptides as a model system. Applied
to a DCL of 81 peptides, our method resolves the full mixture composition from just 13 overlapping peaks, revealing that lateral couplings dominate the assembly energetics.Where the paper says it
Conclusion
The work presents an efficient framework to reconstruct the composition of a complex combinatorial mixture from partial experimental data This approach simplifies library design and enables more practical self-assembly protocols. The method successfully resolves the full composition of a DCL with 81 species from only 13 HPLC peaks. p. 0 We demonstrate
this approach using dynamic combinatorial libraries (DCLs) of peptides as a model system. Applied
to a DCL of 81 peptides, our method resolves the full mixture composition from just 13 overlapping peaks, revealing that lateral couplings dominate the assembly energetics.Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Statistical physics inspired models can be used to express chemical potential as a function of a reduced set of parameters, such as
intrinsic building-block energies and pairwise coupling terms between neighboring building blocks in the sequence.
This allows for characterizing the energy landscape of self-assembled compounds using a small number of parameters. -
The framework can be coupled with the law of mass action to yield equilibrium concentrations across the entire reactive network, enabling the determination of
the full composition of this DCL, from just 13 overlapping peaks
even when standard analytical techniques fail. -
An improved AI system can perform inference by minimizing a distance measure between theoretical predictions and experimental measurements using methods like a random walk minimization or Powell descent gradients to find the set of parameters that maximizes the likelihood with available partial data.
-
The system can be designed to determine the composition of complex reactive mixtures by expressing
the free energy of each compound in the mixture using a reduced number of parameters,
significantly reducing parameter requirements compared to enumerating all possible molecules.
Abstract
Self-assembled mixtures enable the exploration of vast chemical diversity, yet characterizing the hundreds of coexisting species they contain remains a major challenge. Standard analytical techniques such as chromatography and spectroscopy produce overlapping signals that cannot be resolved by conventional deconvolution methods, which neglect the concentration correlations imposed by the reactive network. Here, we introduce a framework that bridges statistical physics and analytical chemistry to deconvolute complex self-assembled mixtures from partial spectral data. Inspired by Potts models of protein folding, we express the chemical potential of each compound as a function of a reduced set of parameters-intrinsic building-block energies and pairwise coupling terms between neighboring building blocks in the sequence. Coupling this model with the law of mass action yields equilibrium concentrations across the entire reactive network. The model parameters can be infered by fitting predicted spectroscopic signals to experimentally measured partial data. We demonstrate this approach using dynamic combinatorial libraries (DCLs) of peptides as a model system. Applied to a DCL of 81 peptides, our method resolves the full mixture composition from just 13 overlapping peaks, revealing that lateral couplings dominate the assembly energetics. Beyond DCLs, this strategy offers a general route for analyzing self-assembled systems from partial data and extends naturally to target-driven identification of high-affinity binders.
Related papers
- A System-Independent Metadynamics Strategy for Reactive Training Data: Application to Gas-Phase Organic Reactions
- A Machine Learning Model for the Chemistry of a Solvated Electron
- Compression of virtual spaces in transcorrelated methods via singular value decomposition: application to the G2 set
- Charge Transfer with a Spin. II: A Framework for Diabatization which Localizes Charge and Spin
- Diabatic Seam Space Sampling for Hydrogen Tunneling Systems with Nuclear-Electronic Orbital Theory
- Augmented Roothaan-Hall Hessian Applied to Spin-Restricted Open-Shell Density-Functional Theory