Towards a General Treatment for Cancer: Part 2— On the History of Drug Design

This Article is the second one in a multi part series summarizing recent findings from a 6 month joint venture between Charité Berlin and RWTH Aachen. Check out my profile for the first part, outlining the history of cancer treatments and underscoring why targeted, patient specific treatments are desirable. If you like my articles please consider following me, and feel free to reach out at martintannhaus@gmail.com. Scientific research is only possible with the support of the public. Thank you for reading, Martin Tannhaus.

How to discover medical drugs?

The practice of discovering therapeutic drugs has evolved throughout history, undergoing drastic changes since its inception. Prior to the modern era, drug discovery was largely “empirical” in the sense that early medicine relied mostly on trial-and-error and the use of natural products derived from plants and minerals, and other biological sources. With the rise of organic chemistry in the late 19th and early 20th centuries, researchers began to isolate and synthesize compounds systematically, exemplified by breakthroughs like Paul Ehrlich’s synthesis of salvarsan for syphilis in 1908 [1], one of the earliest examples of a purpose-designed therapeutic agent rather than an extract of a natural remedy. Scientific advances in structural chemistry also led to the conceptualization of drug-receptor interactions, laying a rational foundation for further progress. Systematic screening of compound libraries and chemical modification gradually became standard practice in medicinal chemistry.

By the mid-20th century, natural product discovery dominated early drug discovery, particularly for antibiotics, analgesics and other classes of drugs. Techniques such as fermentation and whole-cell screening identified bio-active metabolites from microorganisms and plants. While natural products provided complex and biologically active starting points, challenges like compound isolation, targeted structural design, and the synthesis of complex molecules — or rather the difficulty therein — constituted major obstacles that slowed down development significantly. Advances in bioinformatics, proteomics, transcriptomics, and gene expression, among others, aim to utilize the drop in genome sequencing cost of the last decades to accelerate the discovery and development of “natural metabolites”, e.g. metabolic byproducts produced by actinomycetes, and have marked the dawn of microbial genome mining for the discovery of novel medical agents. [2]

The later decades of the 20th century witnessed major shifts in the field. High-throughput screening allowed the rapid testing or large synthetic and natural compound libraries against biological targets, drastically increasing throughput compared to slow manual screening processes. Structure-based drug design, leveraging X-ray crystallography and molecular modeling, enabled rational optimization of candidate compounds based on the structure of three-dimensional receptor binding sites. [3]

Reverse pharmacology (also known as target-based drug discovery) emerged as the dominant standard, focusing on a hypothesized protein target first, and then screening for molecules that may modulate its function. Again, rapid advancements in human- & general genome sequencing accelerated this shift by enabling the rapid cloning and expression analysis of purified targets for biochemical screening. As a downside, target-based approaches sometimes produced compounds with excellent binding but poor efficacy in complex biological systems, and the reliance on already available libraries limited the novelty of possible discoveries. [4]

Despite these technologies, the traditional drug discovery pipeline has been criticized for high attrition rates, long development times, and high costs. Extensive testing from hit identification through clinical trials often results in failure due to lack of efficacy, potential toxicity, or poor pharmacokinetics, consuming substantial time and capital before successful drugs reach the market. Traditional screening also explores only a tiny faction of chemical design space, constraining the discovery of truly novel mechanism-based therapeutics [5]

The advent of the genomic era and omics technologies transformed the landscape again. Genome sequencing, initially through Sanger Sequencing [6] and later through high-throughput next-generation sequencing [7], enabled comprehensive cataloguing of genetic variation across diseases, accelerating target identification and validation. Parallel advances in proteomics allowed researchers to profile the expression, structure, and interactions of proteins, the principal technical architects of cellular function, in a disease context. Integrating genomic and proteomic data has permitted a more holistic understanding of biological pathways and potential intervention points that were previously thought impossible.

These insights have catalyzed the development of de novo drug discovery, which differs from traditional iterative modification of known compounds. Rather than optimizing analogues from existing chemical scaffolds, de novo approaches use computational generative models to design novel molecular structures ab initio, guided by information about target structure and function. When combined with structural proteomics, accurate three-dimensional models of protein targets can be generated, enabling structure-based design of molecules that precisely complement key active sites. Advances in machine learning — including generative models such as diffusion models, and most recently flow matching frameworks like the “Proteina” Foundation Model developed by Nvidia [8] — are increasingly used to explore vast regions of chemical design space and to propose candidate compounds with precise desired structural properties. These methods promise to reduce reliance on large screening libraries and accelerate the generation of candidate molecules tailored specifically to biological targets identified through genomic and proteomic analysis. [9]

The integration of high-throughput sequencing, proteomic profiling, and advanced computational design thus represents a transition from empirical and incremental discovery towards more predictive, efficient, and targeted therapeutic development, with the potential to address historically intractable diseases and highly heterogeneous conditions such as cancer.

If you made it all the way here, a heartfelt thank you from me for reading! My work would not be possible without the continued support of the public and curious readers and colleagues. In the next part of this series of articles I’ll introduce the data formats that are commonly used in proteomics and outline useful computational tools and libraries for the large scale processing of these data.

— Martin Tannhaus.

Sources:

[1] Historical Perspective and Principles of Drug Design (n.d.), Longdom. https://www.longdom.org/open-access/historical-perspective-and-principles-of-drug-design-107515.html

[2] Natural Products in Drug Discovery: Advances and Opportunities (2016), PubMed. https://pubmed.ncbi.nlm.nih.gov/26739136/

[3] Takenaka T (2001), “Classical vs reverse pharmacology in drug discovery”, BJU International, 88 (Suppl 2): 7–10. doi:10.1111/j.1464–410X.2001.00112.x. PMID 11589663.

[4] Lazo JS (2008), “Rear-view mirrors and crystal balls: a brief reflection on drug discovery”, Molecular Interventions, 8 (2): 60–63. doi:10.1124/mi.8.2.1. PMID 18403648

[5] Emerging trends in computational approaches for drug discovery inmolecular biology (2023), GSC Online Press. https://gsconlinepress.com/journals/gscbps/sites/default/files/GSCBPS-2023-0340.pdf

[6] DNA sequencing with chain-terminating inhibitors (1977), PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC431765/

[7] The sequence of sequencers: The history of sequencing DNA (2016), PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC4727787/

[8] NVIDIA Research (n.d.), Proteina — Generative Models for Protein Design. https://research.nvidia.com/labs/genair/proteina/

[9] Recent Advances in Automated Structure-Based De Novo Drug Design (2024), PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC10966644/