Brandon Da Silva

Sweet bystander becomes a villain Patients with pancreatic cancer often have elevated blood levels of CA19-9, a carbohydrate antigen present on many proteins. CA19-9 is thus commonly used as a biomarker for diagnosing and monitoring disease progression. In a study of mice, Engle et al. found that CA19-9 may be more than an innocent bystander that marks the presence of pancreatic disease; it may play a causal role in disease (see the Perspective by Halbrook and Crawford). Transgenic mice expressing the human enzymes that add CA19-9 to proteins developed severe pancreatitis that could be reversed by treatment with CA19-9 antibodies. When the transgenic mice also harbored a Kras oncogene, they went on to develop pancreatic cancer. These unexpected observations suggest new avenues for the treatment of pancreatic disease. Science , this issue p. 1156 ; see also p. 1132

Pancreatic ductal adenocarcinoma (PDA) is a lethal malignancy with limited treatment options. Although metabolic reprogramming is a hallmark of many cancers, including PDA, previous attempts to target metabolic changes therapeutically have been stymied by drug toxicity and tumour cell plasticity. Here, we show that PDA cells engage an eIF4F-dependent translation program that supports redox and central carbon metabolism. Inhibition of the eIF4F subunit, eIF4A, using the synthetic rocaglate CR-1-31-B (CR-31) reduced the viability of PDA organoids relative to their normal counterparts. In vivo, CR-31 suppresses tumour growth and extends survival of genetically-engineered murine models of PDA. Surprisingly, inhibition of eIF4A also induces glutamine reductive carboxylation. As a consequence, combined targeting of eIF4A and glutaminase activity more effectively inhibits PDA cell growth both in vitro and in vivo. Overall, our work demonstrates the importance of eIF4A in translational control of pancreatic tumour metabolism and as a therapeutic target against PDA.

Training deep learning models that generalize well to live deployment is a challenging problem in the financial markets. The challenge arises because of high dimensionality, limited observations, changing data distributions, and a low signal-to-noise ratio. High dimensionality can be dealt with using robust feature selection or dimensionality reduction, but limited observations often result in a model that overfits due to the large parameter space of most deep neural networks. We propose a generative model for financial time series, which allows us to train deep learning models on millions of simulated paths. We show that our generative model is able to create realistic paths that embed the underlying structure of the markets in a way stochastic processes cannot.