Campos de fuerza derivados de la mecánica cuántica para la predicción de especificidad y selectividad en complejos proteína-ligando.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La industria farmacéutica ha buscado de manera constante acelerar y abaratar el descubrimiento de nuevas moléculas activas, apoyándose creciente mente en métodos computacionales para la predicción y validación de compuestos. Entre ellos destacan la estimación de la energía libre de unión absoluta, así como la evaluación de especificidad y selectividad en complejos proteína-ligando como los más desafiantes. Para estos fines se han empleado metodologías de acoplamiento molecular, cálculos de punto final y enfoques termodinámicos basados en transformaciones alquímicas o geométricas, ca paces de describir con alta exactitud los principios físico-químicos que gobiernan la formación de complejos moleculares. En particular, los métodos de perturbación alquímica y de no equilibrio han demostrado gran potencial por su exactitud, capacidad de paralelización y mejor muestreo conformacional. Sin embargo, persisten limitaciones asociadas a la adecuada descripción de interacciones intra- e intermoleculares, usualmente modeladas mediante campos de fuerza clásicos no polarizables.
En este contexto, la parametrización de moléculas pequeñas representa un desafío central, especialmente para la asignación precisa de parámetros no enlazantes en ambientes químicos diversos. Recientes avances, como los promovidos por la iniciativa cooperativa y de ciencia abierta Open Force Field, han aprovechado datos de mecánica cuántica de alto nivel en la generación de mejores campos de fuerza. Por otro lado, metodologías de partición de densidad electrónica desarrollados en nuestro laboratorio, han permitido mejorar la asignación de cargas atómicas y parámetros de Lennard-Jones, obteniendo propiedades atómicas desde primeros principios.
Sobre esta base, en esta tesis se desarrolló la herramienta de código abierto ffparaim, destinada a la derivación automatizada de parámetros no enlazantes mediante el protocolo D-MBIS. Este enfoque incorpora el entorno molecular explícito durante la partición de la densidad electrónica, generando parámetros más representativos para campos de fuerza no polarizables.
Las simulaciones realizadas mostraron mejoras significativas en la predicción de energías libres de unión absoluta en sistemas modelo de referencia. En el conjunto de ligandos del sistema BRD4(1) se obtuvo un error cuadrático medio inferior a 1 kcal·mol−1 en ligandos neutros, mientras que la evaluación de selectividad de bromosporina en distintos bromodominios y el sistema modelo lisozima T4 L99A/M102Q dieron en manifiesto la importancia crítica de factores adicionales, como el cambio conformacional del receptor y la correcta definición de la pose inicial del ligando. Corrigiendo estos factores se logró alcanzar predicciones dentro de la exactitud química.
En conjunto, los resultados validan la utilidad de la herramienta computacional generada en esta tesis y de los parámetros no enlazantes derivados mediante la metodología D-MBIS como una estrategia sólida para aumentar la exactitud de campos de fuerza no polarizables en la predicción de afinidad. Estos sientan las bases para el desarrollo de metodologías de simulaciones de dinámica molecular con campos de fuerza a la medida y estrategias de diseño de fármacos con un enfoque más fundamentado en principios físicos.
The pharmaceutical industry has consistently sought to accelerate and reduce the cost of discovering new active molecules, increasingly relying on computational methods for compound prediction and validation. Among the se, the estimation of absolute binding free energy, as well as the assessment of specificity and selectivity in protein–ligand complexes, stand out as most challenging. To this end, molecular docking, endpoint free energy calculations, and thermodynamic approaches based on alchemical or geometric transfor mations have been employed, as they are capable of accurately describing the physicochemical principles governing the formation of molecular complexes. In particular, alchemical perturbation and nonequilibrium methods have shown great potential due to their accuracy, parallelization capability and improved conformational sampling. However, limitations persist in the proper description of intra- and intermolecular interactions, which are usually modeled using non-polarizable classical force fields. In this context, the parameterization of small molecules represents a central challenge, especially for the precise assignment of nonbonded parameters in diverse chemical environments. Recent advances, such as those promoted by the cooperative and open-science Open Force Field initiative, have leveraged high-level quantum mechanical data in the generation of better force fields. On the other hand, electronic density partitioning methodologies developed in our laboratory have enabled improvements in the assignment of ato mic charges and Lennard-Jones parameters, deriving atomic properties from first principles. On this basis, this thesis developed the open-source tool ffparaim, ai med at the automated derivation of nonbonded parameters through the D MBIS protocol. This approach incorporates the explicit molecular environ ment during electronic density partitioning, generating more representative parameters for non-polarizable force fields. The simulations performed showed significant improvements in the prediction of absolute binding free energies in reference model systems. In the BRD4(1) ligand set, a root-mean-square error below 1 kcal·mol−1 was achieved for neutral ligands, while the evaluation of Bromosporine selectivity across different bromodomains and the model system T4 Lysozyme L99A/M102Q highlighted the critical importance of additional factors, such as receptor conformational changes and the proper definition of the initial ligand pose. By correcting for these factors, predictions within chemical ac curacy were achieved. Taken together, the results validate the usefulness of the computational tool developed in this thesis and of the nonbonded parameters derived th rough the D-MBIS methodology as a solid strategy to enhance the accuracy of non-polarizable force fields in affinity prediction. These findings lay the groundwork for the development of molecular dynamics simulation metho dologies with bespoke force fields and drug design strategies more firmly grounded in physical principles.
The pharmaceutical industry has consistently sought to accelerate and reduce the cost of discovering new active molecules, increasingly relying on computational methods for compound prediction and validation. Among the se, the estimation of absolute binding free energy, as well as the assessment of specificity and selectivity in protein–ligand complexes, stand out as most challenging. To this end, molecular docking, endpoint free energy calculations, and thermodynamic approaches based on alchemical or geometric transfor mations have been employed, as they are capable of accurately describing the physicochemical principles governing the formation of molecular complexes. In particular, alchemical perturbation and nonequilibrium methods have shown great potential due to their accuracy, parallelization capability and improved conformational sampling. However, limitations persist in the proper description of intra- and intermolecular interactions, which are usually modeled using non-polarizable classical force fields. In this context, the parameterization of small molecules represents a central challenge, especially for the precise assignment of nonbonded parameters in diverse chemical environments. Recent advances, such as those promoted by the cooperative and open-science Open Force Field initiative, have leveraged high-level quantum mechanical data in the generation of better force fields. On the other hand, electronic density partitioning methodologies developed in our laboratory have enabled improvements in the assignment of ato mic charges and Lennard-Jones parameters, deriving atomic properties from first principles. On this basis, this thesis developed the open-source tool ffparaim, ai med at the automated derivation of nonbonded parameters through the D MBIS protocol. This approach incorporates the explicit molecular environ ment during electronic density partitioning, generating more representative parameters for non-polarizable force fields. The simulations performed showed significant improvements in the prediction of absolute binding free energies in reference model systems. In the BRD4(1) ligand set, a root-mean-square error below 1 kcal·mol−1 was achieved for neutral ligands, while the evaluation of Bromosporine selectivity across different bromodomains and the model system T4 Lysozyme L99A/M102Q highlighted the critical importance of additional factors, such as receptor conformational changes and the proper definition of the initial ligand pose. By correcting for these factors, predictions within chemical ac curacy were achieved. Taken together, the results validate the usefulness of the computational tool developed in this thesis and of the nonbonded parameters derived th rough the D-MBIS methodology as a solid strategy to enhance the accuracy of non-polarizable force fields in affinity prediction. These findings lay the groundwork for the development of molecular dynamics simulation metho dologies with bespoke force fields and drug design strategies more firmly grounded in physical principles.
Description
Tesis presentada para optar al grado de Doctor en Ciencias con mención en Química.
Keywords
Dinámica molecular, Biología molecular, Simulación por computadores