caMPO: Correlation-aware multiparameter optimization for data-driven compound prioritization in drug design
Abstract
Drug design is inherently a multiparameter optimization (MPO) problem, where several properties need to be simultaneously improved. Beyond activity against the therapeutic target, other relevant properties related to pharmacokinetics and safety are considered during lead optimization. A common MPO approach for compound ranking and prioritization is to identify a set of property objectives and desired ranges (target property profile) and aggregate them into a single MPO score. Sometimes a relative importance or weight is assigned to each property objective. However, this MPO approach suffers from a few drawbacks. For instance, positively correlated objectives may be inadvertently overemphasized while independent objectives may be masked, resulting in a biased approach. In this work, the correlation‑aware MPO (caMPO) framework is introduced to automatically derive weights that account for correlations in the data. This method down‑weights property objectives that are largely predictable from others and up‑weights those that contribute unique information. caMPO was first validated in a controlled setting with synthetic data to assess how predefined correlation patterns affect the derived weights. Moreover, analyses on real drug discovery projects showed that the approach enabled the identification of development candidates (DCs), with an enrichment in compound ranking. For most projects (13/15), DCs were ranked within the top 25% of project compounds, and in approximately half of the projects (7/15) DCs were within the top 10% of the ranking. caMPO consistently matched or exceeded the performance of other standard MPO approaches. Importantly, caMPO can be adapted to project-specific target property profiles, helps to define complex MPO scores while reducing redundancy among property objectives, and gives insights into optimization trade-offs. Taken together, caMPO provides a flexible and data-driven MPO framework for compound prioritization and property optimization.
Related
- Optimization and Evaluation of Polymer Inclusion Membrane (PIM) Containing 10% Copoly Eugenol Divinyl Benzene as Carrier Compound for Malachite Green Transport
- Optimization of compound K extraction using natural deep eutectic solvents and evaluation of mitochondrial protective activity
- Climate change-driven compound dust storm and heatwave extremes in Central Asia under future CMIP6 scenarios
- Correlation-consistent Gaussian basis sets for copper solids from material-constrained atomic optimization
- Optimization of RF Sputtering Power for High-Performance ITZO Thin-Film Transistors: Correlation Between Film Properties, Oxygen Defects and Electrical Performance
BibTeX
@misc{garellick2026campo,
title = {{caMPO: Correlation-aware multiparameter optimization for data-driven compound prioritization in drug design}},
author = {Daniel Garellick and Michele Ghini and Jessica Lanini and Nadine Schneider and Grégori Gerebtzoff and Raquel Rodríguez-Pérez},
howpublished = {ChemRxiv},
year = {2026},
month = aug,
doi = {10.26434/chemrxiv.15008121/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008121/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - caMPO: Correlation-aware multiparameter optimization for data-driven compound prioritization in drug design
AU - Daniel Garellick
AU - Michele Ghini
AU - Jessica Lanini
AU - Nadine Schneider
AU - Grégori Gerebtzoff
AU - Raquel Rodríguez-Pérez
PY - 2026
DA - 2026/08/31
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008121/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008121/v1
AB - Drug design is inherently a multiparameter optimization (MPO) problem, where several properties need to be simultaneously improved. Beyond activity against the therapeutic target, other relevant properties related to pharmacokinetics and safety are considered during lead optimization. A common MPO approach for compound ranking and prioritization is to identify a set of property objectives and desired ranges (target property profile) and aggregate them into a single MPO score. Sometimes a relative importance or weight is assigned to each property objective. However, this MPO approach suffers from a few drawbacks. For instance, positively correlated objectives may be inadvertently overemphasized while independent objectives may be masked, resulting in a biased approach. In this work, the correlation‑aware MPO (caMPO) framework is introduced to automatically derive weights that account for correlations in the data. This method down‑weights property objectives that are largely predictable from others and up‑weights those that contribute unique information. caMPO was first validated in a controlled setting with synthetic data to assess how predefined correlation patterns affect the derived weights. Moreover, analyses on real drug discovery projects showed that the approach enabled the identification of development candidates (DCs), with an enrichment in compound ranking. For most projects (13/15), DCs were ranked within the top 25% of project compounds, and in approximately half of the projects (7/15) DCs were within the top 10% of the ranking. caMPO consistently matched or exceeded the performance of other standard MPO approaches. Importantly, caMPO can be adapted to project-specific target property profiles, helps to define complex MPO scores while reducing redundancy among property objectives, and gives insights into optimization trade-offs. Taken together, caMPO provides a flexible and data-driven MPO framework for compound prioritization and property optimization.
ER -