Highlights d Trained a neural network to predict APA using data from over 3 million reporters d Visualized learned features to reveal a rich cis-regulatory code for APA d Developed and tested an algorithm to accurately engineer polyadenylation signals d Predicted and experimentally characterized over 12,000 human APA variants

All stem cells have the ability to balance their production of self-renewing and differentiating daughter cells. The germline stem cells(GSCs) of the Drosophila ovary maintain such balance through physical attachment to anterior niche cap cells and stereotypic cell division, whereby only one daughter remains attached to the niche. GSCs are attached to cap cells via adherens junctions, which also appear to orient GSC division through capture of the fusome, a germline-specific organizer of mitotic spindles. Here we show that the Rab11 GTPase is required in the ovary to maintain GSC-cap cell junctions and to anchor the fusome to the anterior cortex of the GSC. Thus, rab11-null GSCs detach from niche cap cells, contain displaced fusomes and undergo abnormal cell division, leading to an early arrest of GSC differentiation. Such defects are likely to reflect a role for Rab11 in E-cadherin trafficking as E-cadherin accumulates in Rab11-positive recycling endosomes (REs) and E-cadherin and Armadillo (β-catenin) are both found in reduced amounts on the surface of rab11-null GSCs. The Rab11-positive REs through which E-cadherin transits are tightly associated with the fusome. We propose that this association polarizes the trafficking by Rab11 of E-cadherin and other cargoes toward the anterior cortex of the GSC,thus simultaneously fortifying GSC-niche junctions, fusome localization and asymmetric cell division. These studies bring into focus the important role of membrane trafficking in stem cell biology.

SUMMARY Engineering gene and protein sequences with defined functional properties is a major goal of synthetic biology. Deep neural network models, together with gradient ascent-style optimization, show promise for sequence design. The generated sequences can however get stuck in local minima and often have low diversity. Here, we develop deep exploration networks (DENs), a class of activation-maximizing generative models, which minimize the cost of a neural network fitness predictor by gradient descent. By penalizing any two generated patterns on the basis of a similarity metric, DENs explicitly maximize sequence diversity. To avoid drifting into low-confidence regions of the predictor, we incorporate variational autoencoders to maintain the likelihood ratio of generated sequences. Using DENs, we engineered polyadenylation signals with more than 10-fold higher selection odds than the best gradient ascent-generated patterns, identified splice regulatory sequences predicted to result in highly differential splicing between cell lines, and improved on state-of-the-art results for protein design tasks.

No abstract