Genomic-driven nutritional interventions for radiotherapy-resistant rectal cancer patient

Abstract

Radiotherapy response of rectal cancer patients is dependent on a myriad of molecular mechanisms including response to stress, cell death, and cell metabolism. Modulation of lipid metabolism emerges as a unique strategy to improve radiotherapy outcomes due to its accessibility by bioactive molecules within foods. Even though a few radioresponse modulators have been identified using experimental techniques, trying to experimentally identify all potential modulators is intractable. Here we introduce a machine learning (ML) approach to interrogate the space of bioactive molecules within food for potential modulators of radiotherapy response and provide phytochemically-enriched recipes that encapsulate the benefits of discovered radiotherapy modulators. Potential radioresponse modulators were identified using a genomic-driven network ML approach, metric learning and domain knowledge. Then, recipes from the Recipe1M database were optimized to provide ingredient substitutions maximizing the number of predicted modulators whilst preserving the recipe's culinary attributes. This work provides a pipeline for the design of genomic-driven nutritional interventions to improve outcomes of rectal cancer patients undergoing radiotherapy.

Figure 1

Overview of approach. (A) Network representation of over-expressed proteins (yellow) and biological functions (blue) in RT-resistant RC patients. Over-expressed proteins were experimentally identified by one of the authors of this work. Their corresponding biological functions were extracted from the Gene Ontology’s Biological Processes. Over-expression of all genes but PTEN leads to increased lipid availability resulting in cancer-promoting effects including increased nutrient storage (CDS1, CDS2), activation of stress and response signaling pathways (PLA2G5, ELOVL family, FASN and PLP family), and increased immunosuppressive properties (PTDSS1). (B) Radioresponse modulators identification module. Food protein targets and RT-resistant-associated proteins are mapped onto a multiscale interactome of proteins and biological functions. A biased random walk with restarts (RWR) propagates the effects of food molecules and phenotype, revealing the most affected proteins and biological functions. Top food molecules with the most similar propagated profiles to the phenotype are used to create a list of food ingredients with potentially beneficial RT response modulation activity. (C) Recipe generation module. Using FoodBERT, Recipe1M recipes are optimized to increase the number of ingredients with beneficial radioresponse modulation properties.

Figure 2

The presented model outperforms the baseline approach across all values of restart probability. AUROC of purely random walk based approach and the proposed approach combining the method with metric learning as a function of restart probability on the drug-disease prediction task.

Figure 3

The multiscale interactome identifies proteins and biological functions related to RT modulation. (A) The multiscale interactome involving Mangiferol and RT modulation. Mangiferol modulates RT response by targeting CTNNB1 and TNF which has downstream effects in osteoblast differentiation by being linked to FASN. (B) The multiscale interactome involving Dihydrosphingosine and RT modulation. Dihydrosphingosine modulates RT response primarily by targeting PLPP2, PLPP3 and CERS2, which are linked to the ELOVL family of genes, inhibiting fatty acid biosynthetic and elongation processes.

Figure 4

Ingredient substitutions using FoodBert embeddings and food-chemical information. (A) Visualisation of FoodBert embeddings. 2D representations found using PCA of the 768 dimensional FoodBert embeddings for some ingredients. Ingredients close in this space appear in similar contexts. (B) Example ingredient substitutions. Some example substitutions found by the K-nearest neighbor algorithm in the embedding space and additional filtering to increase the number of beneficial bioactive molecules. (C) Some example ingredient substitutions within popular recipes.