Driving the Discovery of Treatments in Medicine Through Automated Multimodal Machine Learning

Many diseases still have ineffective treatments, resulting in high mortality rates and making the discovery of new therapies essential for humanity. However, developing new therapies is a time-consuming and costly process. Machine learning methods have the potential to accelerate these discoveries by identifying latent knowledge within large volumes of medical data.

Latent knowledge can be understood as relationships or patterns hidden in large datasets. Its discovery involves the automated identification of elements that can provide new insights that would be difficult to obtain using traditional analytical approaches. However, automating this process is often limited by its strong dependence on domain experts — a limitation that Automated Machine Learning (AutoML) methods can help mitigate. Despite their advantages, these methods still face a major challenge: the lack of transparency in the generated models, which represents a significant obstacle, particularly in critical domains such as healthcare.

In this talk, I will present our ongoing research on a multimodal approach that combines state-of-the-art AutoML, Natural Language Processing, and explainability techniques to overcome these limitations and enable faster, more interpretable discoveries in the medical field. By integrating structured data, such as clinical and genetic information, with unstructured data, including scientific literature and medical texts, this approach has the potential to uncover patterns that may accelerate the discovery of new therapies and improve the effectiveness of existing treatments.