Earlier this spring I attended my first in person conference in over two years, the national meeting of the American Chemical Society (ACS). It was immensely enjoyable to reconnect with professional peers and to honor my good friend and mentor Kate Holloway, who last year received the highly prestigious ACS award in pharmaceutical research. I met Kate as a graduate student while attending my first Gordon Research Conference in computer aided-drug design (CADD), where Kate was and still is a fixture. She is somewhat of a celebrity in our small knit scientific community; in 2005 she was the answer to a Jeopardy question in the category “She Invented What” for her work on the design of Crixivan, one of the first marketed HIV protease inhibitors. That’s about as cool as it gets for a bunch of chemistry nerds who sit behind computers all day, designing molecules with the hopes of one day turning them into drugs. Unlike Kate, though, for most of us that is never the case, yet optimism for the field, and the promise it holds to transform how therapies are brought to patients, is higher now than ever before. The field of CADD dates back to the emergence of the personal computer in the late 70s. In 1981 CADD made the cover of Fortunate magazine and was hailed “the next industrial revolution” (right, inset). Flash forward forty years to now, where we’re seeing unprecedented levels of private investment in the space of AI-driven life science companies as well as successful public debuts of technology companies that are building next generation, AI-assisted drug discovery (AIDD) platforms (for example, Recursion, Exscientia) as well as more traditional CADD companies (for example, Schrödinger). While it’s fair to say that the field has matured significantly since the early days of CADD, which now encompasses AI-driven technologies, we owe a debt to scientists like Kate who laid the groundwork decades ago. The leaps made in the field over the last decade are attributable largely to the convergence of three factors: data, algorithms, and cloud computing. High-throughput experimentation coupled with the ability to crunch numbers at a scale once inconceivable enables us to do things today that would have been time and cost prohibitive even in the early 2000s, when I entered this field as a graduate student. Modern algorithms - AI and otherwise - enable us to leverage even sparse data, a problem that still plagues our industry (and warrants its own discussion). The current pandemic has driven home the need to develop therapies expeditiously, a challenge with which the pharmaceutical industry has struggled for decades. In silico methods hold great promise for driving down costs and timelines, improving the manual and costly laboratory-centric processes that define traditional development with more efficient development design-make-test-analyze (DMTA) cycles. We’re starting to see evidence that this promise may be realized, with recent announcements from Recursion, Exscientia, and Insilico Medicine that programs are progressing into clinical development in a fraction of the typical time (see here, here, and here]. Perhaps most strikingly, though, is that we are seeing not only a shift in how drugs are designed, but also where they are designed. A recent study published by the Boston Consulting Group highlighted the near exponential growth of discovery pipelines within AI-native drug discovery companies, painting a stark contrast to the stagnation of the early stage pipeline of big pharma . Increasingly CADD and AIDD companies are taking matters into their own hands, creating assets with their platforms rather than simply enabling big pharma and biotech by licensing their tools (or doing both). So, with all the recent progress the big question is: are we there yet? What is hype versus reality? Can we crank a machine and turn out a drug? In my humble opinion, the answer is no. And that is for a variety of reasons, mainly (1) biology is incredibly complex and our understanding of how biology relates to disease is limited (2) many components of the drug discovery process are not optimized for speed and scale, and (3) we still lack enough high quality and high quantity data for AI to be utilized to its full potential. A perspective published by Andreas Bender and Isidro Cortés-Ciriano last year nicely sums its up : “In short, AI in drug discovery needs quantitative variables and labels that are meaningful, but we are often insufficiently able to determine which variables matter, to define them experimentally (and on a large enough scale) and to label the biology for AI to succeed on a level that is compatible with the current investment and hope in the area.” In other words, while (arguably) traditional CADD approaches have made it to the “plateau of productivity” stage, AI in drug discovery is still at the “peak of inflated expectations” stage of the hype cycle. But reasons to be optimistic abound. Advances in automated, scalable, and reliable chemical synthesis and experimental testing hold great promise to address some of the challenges associated with data generation and data quality. As Peter Diamondis asserts in his book The Future is Faster than You Think, the convergence of technology and AI combined with human intelligence is what will be required to take giant leaps forward. Until then the impact that in silico approaches has on drug discovery will remain limited. However, as a hopeful skeptic, I remain encouraged that one day as an industry we will impact the design of all drugs with code. And then we’ll get back on the Jeopardy board. References: Jayatunga M et al. AI in small molecule drug discovery: a coming wave? Nat Rev Drug Discov. 2022. 21: 175-176 Bender A and Cortés-Ciriano I. Artificial intelligence in drug discovery: what is realistic, what are illusions? Part 1: Ways to make an impact, and why we are not there yet. Drug Discov Today. 2021. 26(2): 511-524.