Building in silico models to predict chemical properties and activities is a crucial step in drug discovery. However, drug discovery projects are often characterized by limited labeled data, hindering the applications of deep learning in this setting. Meanwhile advances in meta-learning have enabled state-of-the-art performances in few-shot learning benchmarks, naturally prompting the question: Can meta-learning improve deep learning performance in low-resource drug discovery projects? In this work, we assess the efficiency of the Model-Agnostic Meta-Learning (MAML) algorithm - along with its variants FO-MAML and ANIL - at learning to predict chemical properties and activities. Using the ChEMBL20 dataset to emulate low-resource settings, our benchmark shows that meta-initializations perform comparably to or outperform multi-task pre-training baselines on 16 out of 20 in-distribution tasks and on all out-of-distribution tasks, providing an average improvement in AUPRC of 7.2% and 14.9% respectively. Finally, we observe that meta-initializations consistently result in the best performing models across fine-tuning sets with $k \in \{16, 32, 64, 128, 256\}$ instances.

In startup world these days, the word "biotech" is increasingly accompanied by "computational" and two, two-letter initialisms: AI and ML. Those tools--artificial intelligence and machine learning, respectively--have been around for decades, but in recent years have become faster and cheaper, accelerating their use by those in the business of discovering and developing new drugs. Another startup looking to take advantage of those improvements, South San Francisco-based Genesis Therapeutics, has scored $4.1 million in seed funding and publicly joined the growing fray of biotechs with grand ambitions of disrupting the slow, costly process of discovering and developing new medicines. Andreessen Horowitz, also known as a16z, led its seed round, one of a handful of seed-stage investments it has made in biotech. Felicis Ventures, another VC firm based in Silicon Valley, also invested.

In this episode, Dr. Eric Topol, cardiologist and chair of innovative medicine at Scripps Research, and a16z's general partner on the Bio Fund Vijay Pande, have a conversation around Topol's new book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. What is the impact AI will have on how your doctor engages with you? How will AI impact not just doctor-patient interactions, but diagnosis, prevention, prediction, medical education, and everything in between? Topol and Pande discuss how AI's capabilities for deep phenotyping will shift our thinking from population health to understanding the medical health essence of you, how the industry might respond and the challenges in integrating and introducing the technology into today's system--and ultimately, what that the doctor's visit of the future might look like.

A type of machine learning that works well with small data sets holds promise for drug discovery and development. This methodology could be a useful tool for other areas of chemical research. One-shot learning, a kind of deep learning, differs from other machine-learning approaches in the amount of Vijay Pande. Credit: L.A. Cicerodata required to arrive at a solution. Most applications of machine learning, like image recognition, rely on training a set of algorithms with thousands to trillions of data points. One-shot learning can succeed with hundreds of data points.

Artificially intelligent algorithms can learn to identify amazingly subtle information, enabling them to distinguish between people in photos or to screen medical images as well as a doctor. But in most cases their ability to perform such feats relies on training that involves thousands to trillions of data points. This means artificial intelligence doesn't work all that well in situations where there is very little data, such as drug development. Vijay Pande, professor of chemistry at Stanford University, and his students thought that a fairly new kind of deep learning, called one-shot learning, that requires only a small number of data points might be a solution to that low-data problem. Stanford chemistry Professor Vijay Pande and his students see a future for machine learning in the early stages of drug development.