Diagrammatic view of drug response variability in Africa

We have kick-started a project with H3D to identify genetic traits specific to African populations that determine variability in malaria and tuberculosis treatment

One thing I couldn’t anticipate was my sudden loss of interest in large language models. As an AI practitioner, the advent of astonishing tools like GPT-4, capable of generating convincing text in response to any prompt, should be a cause for celebration. Instead, I find these tools disenchanting and even depressing. With so much text already out there, why do we need to produce more in bulk? We already live in a world where nobody reads. I’m not thrilled about a world where, on top of this, nobody writes. On a personal note, I think this loss of interest stems from an ambition I had six years ago, when I tried to write a fiction book for an entire year assisted by a rudimentary (DIY) AI model fed with works from the literary heroes of my youth. Today, this task is easy to accomplish and therefore boring, non-evocative. I’m unsure how literature will respond to the advent of AI text generators — probably with abstraction, much like painting responded to the advent of photography.

The comparison with visual arts is fitting since, along with my disenchantment with large language models as tools for expressing human knowledge (in culture and science), I’ve regained interest in diagrammatic representations. It is still not easy to generate good diagrams with AI, so it is (for now) a safe space for human craft. While browsing the web, I came across the following diagram by Alfred Barr, from an early MoMA exhibition, depicting the paths that led to cubism and beyond. Representing the history of modern art in a diagrammatic form seems sensible, as it is at least as effective as text, with Barr’s diagram potentially requiring hundreds or even thousands of pages to explain in words. Unfortunately, in science, diagrams often lack aesthetic appeal — to a point where I often believe that scientific clarity and pictorial beauty are mutually exclusive traits. Instead, coming from an art academic, Barr’s handwritten diagrams possess a unique aesthetic boldness, almost compelling you to care about the topic, even if you have no interest in cubism whatsoever.

Diagrammatic view of the history of cubism, by Alfred Barr. “Cubism and Abstract Art”, 1936.

I think it is the aesthetic boldness what gives the Barr diagram the impression of solid truth. This can be misleading and dangerous, more so in this case, where a Europe-centric view of modern art is offered. As it turns out, anticolonial derivatives of Barr’s diagram exist to illustrate, for example, Belgian exploitation of the Congo and how this has influenced modern art:

Anticolonial view of the history of Abstract Art, by Hank Willis Thomas, source: MoMA Magazine.

For quite some time, I have been working on a diagram to represent the effects of medicines in our bodies, only to discover that the same biases present in Barr’s scheme also apply to our approach to drug treatment. Historically, drug development has been predominantly driven by the Western world, which often leads to research that overlooks the needs and genetic diversity of other populations. Consequently, this can result in medicines that are less effective or have adverse effects for certain ethnic groups. Unsurprisingly, the most underrepresented groups tend to be the populations of the Global South.

H3D-Ersilia collaboration to capture African genetic diversity in drug treament models. Source: Ersilia

Recently, thanks to the support of a GRADIENT grant, we have established a collaboration with the H3D Centre to identify new genetic traits specific to African populations that may determine variability in antimalarial and antituberculosis drug responses. The diagram above summarizes the scope of the project. You can follow our progress in this GitHub repository. Admittedly, it is a complex scheme, but I hope it is clear that genetic factors, particularly those of African ethnicities, are explicitly represented at the core of the figure. The goal is to include gene variation as a central component of our data-driven AI models for pharmacology, thus emphasizing the importance of geographical diversity as we develop novel therapies. As a small non-profit organization contributing to science in the Global South, this is undoubtedly our most ambitious endeavor, and we are extremely excited about it.

I am cautiously experimenting with GPT-4. This AI tool has offered grammatical corrections and intermittent suggestions throughout the text, encompassing this very disclaimer note.