Darryl Russell of the University of Adelaide in Australia is seeking safer contraceptives that block ovulation without altering hormone levels and cause fewer side effects using an automated in-vitro screening platform that measures cell adhesion in the cumulus-oocyte complex, which is required to release the oocyte from the ovary. In Phase I, they built the screening platform by isolating cumulus-oocyte complexes from mice, culturing them in fibronectin-coated multi-well plates, and quantifying adhesion in a 96-well plate format using an automated assay. In a first run, they screened a library of 129 FDA-approved chemical compounds over four months and identified seven candidate contraceptives with known protein targets, one of which showed a strong reduction in ovulation when tested in mice. In Phase II, they will study whether one of the main target proteins identified in their first screen is a key target for blocking ovulation. They will also test whether the other candidates from their first screen can block ovulation in mice and screen larger and more diverse libraries to identify new candidate contraceptives and prioritize them for further drug development and testing.

Drew Arenth, Benjamin Fels, and Suvrit Sra of Macro-Eyes in the U.S. are applying a statistical machine learning approach to the immunization supply chains of health facilities in Tanzania that accurately and continuously predicts demand to ensure the right vaccines and levels are being stocked. Currently, vaccine supply is largely fixed or driven by depleted stocks. This leaves children unable to be vaccinated due to stock outs at clinics, as well as often high levels of waste, which could both be overcome by better forecasting vaccination needs for individual clinics. In Phase I, they worked with an NGO and the Ministry of Health to access routinely-collected daily vaccination data from 710 health facilities spread across Tanzania. These data were then used to train algorithms to identify predictive patterns that were tested on independent datasets. This led to a model that could accurately forecast future vaccine consumption. In Phase II, they will work out how best to integrate their approach as an automated component within the existing supply chain infrastructure in Tanzania and develop tools and train advocates to demonstrate its value and encourage implementation and adoption.

Peter Wagstaff of Self Help Africa in Ireland will build an advanced machine learning algorithm that automatically analyzes high-resolution satellite images for near real-time, low-cost detection of crop pests and diseases across wide, varied landscapes. Current detection methods are either resource- or cost-intensive and limited in their ability to provide up-to-date information across large and complex geographic areas. Crop pests and diseases can alter leaf color and expose soil, which can be detected by very high-resolution satellite imaging. They will combine satellite images provided by their partner with field data on the fall armyworm crop pest collected by their project team over 18 months on smallholder plots in the Balaka district in Malawi. These data will be used to train an algorithm to detect pests and diseases. They will use cloud-based workflows to enable computationally intensive processing of large quantities of high-resolution images in near real-time. The accuracy of the algorithm will be evaluated by an independent field survey. Note: This grant is funded by the Foundation for Food and Agriculture Research (FFAR).