A carregar...
Publicação
Benchmarking Computational Algorithms for Enhanced Drug Discovery - Insights from Clarivate’s pre-competitive algorithm benchmarking consortium
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.57 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O desenvolvimento de medicamentos é um processo complexo que requer a integração de informações de muitas áreas diferentes do conhecimento. Durante a fase pré-clínica, dados ómicos são frequentemente utilizados para esclarecer as alterações fenotípicas causadas pela exposição a um agente terapêutico, permitindo assim a reconstrução da cascata bioquímica responsável pelo efeito farmacológico (frequentemente chamada de reconstrução do mecanismo de ação). Esta etapa é crucial para caracterizar um agente terapêutico, e os investigadores dependem de inúmeras ferramentas computacionais para este processo. No entanto, a abundância de soluções disponíveis e a falta de estruturas de avaliação padronizadas representam desafios na identificação de ferramentas confiáveis e eficientes. Com isso em mente, a Clarivate lidera o Algorithm Benchmarking Consortium (ABC), um consórcio baseado em subscrição para farmacêuticas que avalia ferramentas computacionais para vários casos de estudo. Dentro do ABC, 27 algoritmos (13 baseados em topologia, 6 baseados em conectividade e 8 baseados em enriquecimento) para identificar reguladores-chave e inferir proteínas de sinalização desreguladas a partir de dados transcriptómicos foram comparados para fornecer uma avaliação imparcial dos métodos disponíveis. Os algoritmos foram avaliados juntamente com dados transcriptómicos derivados de 7 bases de dados e redes moleculares da OmniPath e MetaBaseTM. As métricas de desempenho incluíram precisão na previsão de reguladores causais, escalabilidade computacional e robustez em relação às variações dos dados utilizados pelos algoritmos. Os algoritmos de raciocínio causal, randomWalk, e não causal, overconnectivity, apresentaram resultados superiores em geral para a recuperação direta de alvos. Entre as ferramentas baseadas em enriquecimento, o wmean foi o que apresentou melhor desempenho. A nível de dados de referência, a rede molecular MetaBase apresentou melhores resultados. E a nível de dados de consulta, as assinaturas derivadas de tecidos superaram as derivadas de linhas celulares. Através deste caso de uso, o ABC enfatiza a importância da reconstrução do mecanismo de ação de uma perturbação para obter uma melhor compreensão dos efeitos farmacológicos dos agentes terapêuticos. Em última análise, isso é essencial para caracterizar perfis de segurança, identificar novas indicações (reposicionamento de medicamentos) e explorar estratégias combinatórias potenciais com efeitos sinérgicos.
Drug development is a complex process that requires integrating information from many different fields of knowledge. During the preclinical phase, omics measurements are often used to clarify the phenotypical changes caused by exposure to a therapeutic agent, thereby enabling the reconstruction of the biochemical cascade responsible for the pharmacologic effect (often called Mechanism of Action (MoA) reconstruction). This step is crucial in characterizing a therapeutic agent, and researchers rely on numerous computational tools for this process. However, the abundance of available solutions and the lack of standardized assessment frameworks pose challenges in identifying reliable and efficient tools. To address this, Clarivate leads the pre-competitive subscription-based Algorithm Benchmarking Consortium (ABC), which evaluates computational tools for various use cases. Within the ABC framework, 27 algorithms (13 topology-based, 6 connectivity-based, and 8 enrichment-based algorithms) for identifying key regulators and inferring dysregulated signalling proteins from transcriptomics data were benchmarked to provide an unbiased assessment of available methods. The algorithms were evaluated together with transcriptomic data derived from 7 databases and molecular networks from OmniPath and MetaBaseTM. Performance metrics included accuracy in predicting causal regulators, computational scalability, and robustness to input data variations. Causal reasoning, randomWalk, and non-causal, overconnectivity, algorithms showed superior results overall for direct target recovery. Among enrichment-based tools, wmean was the best performer. At the reference dataset level, MetaBase network led to the best outcomes. At the query dataset level, tissue-derived signatures outperformed cell line-derived ones. Through this use case, ABC emphasizes the importance of MoA reconstruction as a means to better understand the pharmacological effects of therapeutic agents. Ultimately, this is essential for characterizing safety profiles, identifying new indications (drug repurposing), and exploring potential combinatorial strategies with synergistic effects.
Drug development is a complex process that requires integrating information from many different fields of knowledge. During the preclinical phase, omics measurements are often used to clarify the phenotypical changes caused by exposure to a therapeutic agent, thereby enabling the reconstruction of the biochemical cascade responsible for the pharmacologic effect (often called Mechanism of Action (MoA) reconstruction). This step is crucial in characterizing a therapeutic agent, and researchers rely on numerous computational tools for this process. However, the abundance of available solutions and the lack of standardized assessment frameworks pose challenges in identifying reliable and efficient tools. To address this, Clarivate leads the pre-competitive subscription-based Algorithm Benchmarking Consortium (ABC), which evaluates computational tools for various use cases. Within the ABC framework, 27 algorithms (13 topology-based, 6 connectivity-based, and 8 enrichment-based algorithms) for identifying key regulators and inferring dysregulated signalling proteins from transcriptomics data were benchmarked to provide an unbiased assessment of available methods. The algorithms were evaluated together with transcriptomic data derived from 7 databases and molecular networks from OmniPath and MetaBaseTM. Performance metrics included accuracy in predicting causal regulators, computational scalability, and robustness to input data variations. Causal reasoning, randomWalk, and non-causal, overconnectivity, algorithms showed superior results overall for direct target recovery. Among enrichment-based tools, wmean was the best performer. At the reference dataset level, MetaBase network led to the best outcomes. At the query dataset level, tissue-derived signatures outperformed cell line-derived ones. Through this use case, ABC emphasizes the importance of MoA reconstruction as a means to better understand the pharmacological effects of therapeutic agents. Ultimately, this is essential for characterizing safety profiles, identifying new indications (drug repurposing), and exploring potential combinatorial strategies with synergistic effects.
Descrição
Palavras-chave
Mechanism of Action Transcriptomics Algorithm benchmarking Causal reasoning