Sperduti, A. | bioRxiv Science

bioRxiv2024

TumFlow: An AI Model for Predicting New Anticancer Molecules

MotivationMelanoma is a severe form of skin cancer increasing globally with about 324.000 cases in 2020, making it the fifth most common cancer in the United States. Conventional drug discovery methods face limitations due to the inherently time consuming and costly. However, the emergence of artificial intelligence (AI) has opened up new possibilities. AI models can effectively simulate and evaluate the properties of a vast number of potential drug candidates, substantially reducing the time and resources required by traditional drug discovery processes. In this context, the development of AI normalizing flow models, employing machine learning techniques to create new molecular structures, holds great promise for accelerating the discovery of effective anticancer therapies. ResultsThis manuscript introduces a novel AI model, named TumFlow, aimed at generating new molecular entities with potential therapeutic value in cancer treatment. It has been trained on the comprehensive NCI-60 dataset, encompassing thousands of molecules tested across 60 tumour cell lines, with a specific emphasis on the melanoma SK-MEL-28 cell line. The model successfully generated new molecules with predicted improved efficacy in inhibiting tumour growth while being synthetically feasible. This represents a significant advancement over conventional generative models, which often produce molecules that are challenging or impossible to synthesize. Furthermore, TumFlow has also been utilized to optimize molecules known for their efficacy in clinical melanoma treatments. This led to the creation of novel molecules with a predicted enhanced likelihood of effectiveness against melanoma, currently undocumented on PubChem. Availability and Implementationhttps://github.com/drigoni/TumFlow. Supplementary informationUploaded.

bioinformatics↗

bioRxiv2023

Improved detection of methylation in ancient DNA

Reconstructing premortem DNA methylation levels in ancient DNA (aDNA) has led to breakthrough studies such as the prediction of anatomical features of the Denisovan, as well as the castration status of ancient horses. These studies relied on computationally inferring methylation levels from damage signals in naturally deaminated cytosines. Because of statistical constraints, this inference requires high-coverage sequencing, and is thus not only expensive but also restricted to samples with exceptional DNA preservation. Instead, a method to directly measure methylation levels in aDNA, as exists in modern DNA samples, would open the door to a more thorough and cost effective ability to study ancient DNA methylation. We have tested two methods for direct methylation measurement developed for modern DNA based on either bisulfite or enzymatic methylation treatments. We find that both methods preserve sufficient DNA yields to allow for methylation measurement. Bisulfite treatment, combined with a single stranded library preparation, shows the least reduction in DNA yields compared to no methylation treatment, as well as the least biases during methylation conversion. In addition, we show that applying bisulfite treatment to [~]0.4-fold coverage sample provides a methylation signal that is comparable to, or even better, than the computationally inferred one. We thus present a method to directly measure methylation in ancient DNA that is cost effective and can be used on a wide variety of ancient samples.

genomics↗