Publicação
Development of a computacional approach to predict the activity of new, lead-like, Kv modulators for chronic pain therapy
| Resumo: | Voltage-gated K+ (Kv) channels are responsible for membrane repolarization following an action potential and for setting the neuronal firing pattern. These channels are key players in several auto-immune diseases. These conditions include multiple sclerosis, rheumatoid arthritis and psoriasis, which are known to heavily impact the life of patients. It is also known that Kv channels play an important part in chronic pain syndrome. Chronic pain is defined as pain that outlives its protective role. It has nefarious socioeconomic consequences that extend farther than just the amount of money spent each year (over $500 billion dollars). The current therapies available for these physiopathologies are either ineffective or have serious side-effects. The goal of this project was to develop a QSAR model capable of predicting the activity of new compounds towards Kv channels, allowing the discovery of new lead-like Kv modulators which could be developed into new drugs to treat autoimmune diseases and for chronic pain therapy. Nowadays, computational approaches have been used to develop drugs targeting several diseases. The field of chemoinformatics can help decreasing the time and money invested into finding new lead-like compounds for a given target by, for example, creating QSAR models to virtually screen large chemical libraries. These models can also supply information which can be useful for compound optimization into a new drug. QSAR models derive mathematical structure-activity relationships from complex data sets and use these relationships to predict the activity of new compounds towards a given target. A two-step classification strategy was undertaken to build the models used in the virtual screening. Four different machine learning (ML) techniques were explored to build the models throughout this dissertation (Random Forest, Support Vector Machines, Multilayer Perceptron, k-Nearest Neighbor). However, the best models were built using the Random Forest ML technique. Several structural and fingerprint descriptor sets were also explored. The Kv modulators data set used to build the models comprised 340 compounds, clustered in 10 different structural classes. The first model, A1, was designed to predict two activity categories: IREL, irrelevant, corresponding to ineffective Kv modulators, and RELV, relevant, corresponding to the effective Kv modulators. The second model, A2, predicted two other activity categories: HRELV, high relevance, which refers to the highly effective Kv modulators, and LRELV, low relevance, which corresponds to the moderately effective Kv modulators. The A2 model was built to allow prioritization of the most promising Kv modulators found within the RELV category. The A1 model was built using CDK ExFP and CDK 3D descriptors and the A2 model was built using CDK GraphOnly and 3D CDK descriptors. These models were used to screen five natural products and two approved-drugs databases, comprising a total of 111179 and 2635 molecules respectively. Following the virtual screening, 20 compounds were selected for experimental validation according to their probability of belonging to the HRELV category (PHRELV). One approved drug, roxithromycin, and one natural product, compound A, were tested for their ability to affect outward K+ currents evoked in acutely isolated small diameter DRG neurons at 1000 nM concentration. Compound A caused alterations in the inactivation profiles of K+ currents, indicating that it is a new Kv modulator. Neither this compound nor its scaffold have been described as Kv modulators. As such, this new scaffold can lead to new autoimmune and chronic pain studies. |
|---|---|
| Autores principais: | Pereira, Gilberto Paulo |
| Assunto: | Modelos QSAR Random Forest Screening virtual Kv Técnicas de aprendizagem automática Teses de mestrado - 2018 |
| Ano: | 2018 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Voltage-gated K+ (Kv) channels are responsible for membrane repolarization following an action potential and for setting the neuronal firing pattern. These channels are key players in several auto-immune diseases. These conditions include multiple sclerosis, rheumatoid arthritis and psoriasis, which are known to heavily impact the life of patients. It is also known that Kv channels play an important part in chronic pain syndrome. Chronic pain is defined as pain that outlives its protective role. It has nefarious socioeconomic consequences that extend farther than just the amount of money spent each year (over $500 billion dollars). The current therapies available for these physiopathologies are either ineffective or have serious side-effects. The goal of this project was to develop a QSAR model capable of predicting the activity of new compounds towards Kv channels, allowing the discovery of new lead-like Kv modulators which could be developed into new drugs to treat autoimmune diseases and for chronic pain therapy. Nowadays, computational approaches have been used to develop drugs targeting several diseases. The field of chemoinformatics can help decreasing the time and money invested into finding new lead-like compounds for a given target by, for example, creating QSAR models to virtually screen large chemical libraries. These models can also supply information which can be useful for compound optimization into a new drug. QSAR models derive mathematical structure-activity relationships from complex data sets and use these relationships to predict the activity of new compounds towards a given target. A two-step classification strategy was undertaken to build the models used in the virtual screening. Four different machine learning (ML) techniques were explored to build the models throughout this dissertation (Random Forest, Support Vector Machines, Multilayer Perceptron, k-Nearest Neighbor). However, the best models were built using the Random Forest ML technique. Several structural and fingerprint descriptor sets were also explored. The Kv modulators data set used to build the models comprised 340 compounds, clustered in 10 different structural classes. The first model, A1, was designed to predict two activity categories: IREL, irrelevant, corresponding to ineffective Kv modulators, and RELV, relevant, corresponding to the effective Kv modulators. The second model, A2, predicted two other activity categories: HRELV, high relevance, which refers to the highly effective Kv modulators, and LRELV, low relevance, which corresponds to the moderately effective Kv modulators. The A2 model was built to allow prioritization of the most promising Kv modulators found within the RELV category. The A1 model was built using CDK ExFP and CDK 3D descriptors and the A2 model was built using CDK GraphOnly and 3D CDK descriptors. These models were used to screen five natural products and two approved-drugs databases, comprising a total of 111179 and 2635 molecules respectively. Following the virtual screening, 20 compounds were selected for experimental validation according to their probability of belonging to the HRELV category (PHRELV). One approved drug, roxithromycin, and one natural product, compound A, were tested for their ability to affect outward K+ currents evoked in acutely isolated small diameter DRG neurons at 1000 nM concentration. Compound A caused alterations in the inactivation profiles of K+ currents, indicating that it is a new Kv modulator. Neither this compound nor its scaffold have been described as Kv modulators. As such, this new scaffold can lead to new autoimmune and chronic pain studies. |
|---|