Biology Atlas

Map / The cascade

step · intermediate · partly written

Medicinal chemistry: potency, selectivity, ADMET

One step of the screening cascade walk-through, from disease to clinic.

What is done

A loop: edit a molecule slightly, measure potency, selectivity and properties, keep what helps. The named tactics are matched molecular pairs (collecting what a single change usually does), swapping groups for others with similar behaviour, blocking a metabolic soft spot, and locking a flexible group into a ring; scaffold hopping and property rules are the other named moves. In a measured set of 257 single-atom edits across 18 leads, 29 (11.3%) improved potency 10-fold or more, usually with worse in vitro pharmacokinetics, so most edits do not help. How many compounds a campaign makes is not measured in anything opened: the authors of that paper say "many hundreds, occasionally thousands", citing sources not opened. A study of 156 clinical candidates from one journal over four years found most were structurally distant from the original hit; a company reconstruction covered about 3,000 series (abstracts read). 12345

Reported, not yet replicated

Costs

No source opened prices a lead-optimisation campaign or counts its compounds, so no cost is written. Assay costs for the steps inside the loop are on the linked assay records.

Proves

That a chosen compound meets the programme's own criteria on potency, selectivity against the panel it was tested on, measured exposure and animal efficacy. No standard fixes thresholds for hit, lead or candidate; we found none in the sources opened. 14

Our inference

Cannot prove

That the compound is selective beyond its panel: ten cancer drugs still killed cells after CRISPR deletion of the supposed target, and 243 clinical kinase drugs showed widespread polypharmacology including non-kinase targets. That a potency difference is real: repeat measurements of one compound differ by roughly half a log unit and the same-compound IC50 pairs in public data often differ by more than 0.3 log. That a cell result means what it seems: efflux in the host cell can make a potent enzyme binder read as weak. That the window is wide enough in people. 678910

Reported, not yet replicated

Assays

Regulators require

No regulation read here says how a candidate must be chosen. The ICH question-and-answer text on QT risk sets no numeric hERG margin, floors the free fraction at 1% and asks sponsors to compare against reference drugs run in the same protocol; the first required filing is the IND (21 CFR 312.23(a)(8)), whose pharmacology and toxicology section is built after this step. Our reading. 1112

Our inference

ML role

Does: in a blind challenge on a coronavirus protease (381 submissions, 66 participants), machine-learning methods beat physics-based free-energy methods for potency across the board; one programme and one organiser, and the authors say a matched-pair task might favour physics. A vendor benchmark of free-energy calculation gave a pairwise RMSE of about 1.25 kcal/mol on 1,237 compounds, against about 0.91 kcal/mol between experimental assays. A generative-design paper reported four biochemically active DDR1 inhibitors in 21 days, two active in cells (company-authored; abstract read). Cannot: on activity cliffs, 24 methods on 30 targets all struggled and a fingerprint model beat deep models; a clinical analysis put AI-native molecules at 80 to 90% phase I success and about 40% in phase II, comparable to history, from a small sample (abstract; consultancy authors). Scores against noisy labels cannot honestly beat the noise; no independent prospective efficacy win for AI-designed molecules was found. 13141516178

Reported, not yet replicated

Reports

Scenarios

Modality notes

For nucleic-acid drugs the loop edits sequence and chemistry

Potency becomes how much RNA is destroyed or how much splicing shifts per tissue exposure; selectivity becomes sequence uniqueness; toxicity is partly a property of the backbone. For siRNA the 2018 modification study found the gain in potency and duration was mostly metabolic stability. 181920

See: Antisense oligonucleotide (ASO), siRNA

Our inference

For antibodies selectivity is for the human protein and cell type

Animal safety testing can mislead: the 2006 TGN1412 cytokine storm was not predicted. 2122

See: Therapeutic monoclonal antibodies, TGN1412 (anti-CD28 superagonist antibody)

Reported, not yet replicated

Still open here

Linked from

Sources

  1. Xu, Xinyu, Mailhot, Olivier, Correy, Galen J., et al.. Development of a random background to understand ligand optimization. Nature (2026). paper; opened 2026-11-17 all records citing this
  2. Leach, Andrew G., Jones, Huw D., Cosgrove, David A., et al.. Matched Molecular Pairs as a Guide in the Optimization of Pharmaceutical Properties; a Study of Aqueous Solubility, Plasma Protein Binding and Oral Exposure. Journal of Medicinal Chemistry (2006). paper; opened 2026-11-17 all records citing this
  3. Leeson, Paul D., Springthorpe, Brian. The influence of drug-like concepts on decision-making in medicinal chemistry. Nature Reviews Drug Discovery (2007). paper; opened 2026-11-17 all records citing this
  4. Brown, Dean G.. An Analysis of Successful Hit-to-Clinical Candidate Pairs. Journal of Medicinal Chemistry (2023). paper; opened 2026-11-17 all records citing this
  5. Beckers, Maximilian, Fechner, Nikolas, Stiefl, Nikolaus. 25 Years of Small-Molecule Optimization at Novartis: A Retrospective Analysis of Chemical Series Evolution. Journal of Chemical Information and Modeling (2022). paper; opened 2026-11-17 all records citing this
  6. Lin, Ann, Giuliano, Christopher J., Palladino, Ann, et al.. Off-target toxicity is a common mechanism of action of cancer drugs undergoing clinical trials. Science Translational Medicine (2019). paper; opened 2026-11-17 all records citing this
  7. Klaeger, Susan, Heinzlmeir, Stephanie, Wilhelm, Mathias, et al.. The target landscape of clinical kinase drugs. Science (2017). paper; opened 2026-11-17 all records citing this
  8. Kramer, Christian, Kalliokoski, Tuomo, Gedeck, Peter, et al.. The Experimental Uncertainty of Heterogeneous Public Ki Data. Journal of Medicinal Chemistry (2012). paper; opened 2026-11-17 all records citing this
  9. Landrum, Gregory A., Riniker, Sereina. Combining IC50 or Ki Values from Different Sources Is a Source of Significant Noise. Journal of Chemical Information and Modeling (2024). paper; opened 2026-11-17 all records citing this
  10. Boras, Britton, Jones, Rhys M., Anson, Brandon J., et al.. Preclinical characterization of an intravenous coronavirus 3CL protease inhibitor for the potential treatment of COVID19. Nature Communications (2021). paper; opened 2026-11-17 all records citing this
  11. International Council for Harmonisation. ICH E14/S7B Implementation Working Group: Questions and Answers (Step 4, 21 February 2022). ICH database (2022). regulatory, not peer reviewed; opened 2026-11-17 all records citing this
  12. US Food and Drug Administration. IND content and format (21 CFR 312.23). Electronic Code of Federal Regulations, Title 21, Part 312 (2026). regulatory, not peer reviewed; opened 2026-11-17 all records citing this
  13. MacDermott-Opeskin, Hugo, Scheen, Jenke, Wognum, Cas, et al.. A Computational Community Blind Challenge on Pan-Coronavirus Drug Discovery Data. Journal of Chemical Information and Modeling (2026). paper; opened 2026-11-17 all records citing this
  14. Ross, Gregory A., Lu, Chao, Scarabelli, Guido, et al.. The maximal and current accuracy of rigorous protein-ligand binding free energy calculations. Communications Chemistry (2023). paper; opened 2026-11-17 all records citing this
  15. Zhavoronkov, Alex, Ivanenkov, Yan A., Aliper, Alex, et al.. Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nature Biotechnology (2019). paper; opened 2026-11-17 all records citing this
  16. van Tilborg, Derek, Alenicheva, Alisa, Grisoni, Francesca. Exposing the Limitations of Molecular Machine Learning with Activity Cliffs. Journal of Chemical Information and Modeling (2022). paper; opened 2026-11-17 all records citing this
  17. KP Jayatunga, Madura, Ayers, Margaret, Bruens, Lotte, et al.. How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discovery Today (2024). paper; opened 2026-11-17 all records citing this
  18. Egli, M., Manoharan, M.. Chemistry, structure and function of approved oligonucleotide therapeutics. Nucleic Acids Research (2023). review; opened 2026-11-17 all records citing this
  19. U.S. Food and Drug Administration (label text supplied by the sponsor). SPINRAZA (nusinersen) injection, for intrathecal use: US prescribing information (initial US approval 2016, revised 12/2016). FDA Drugs@FDA label archive (2016). regulatory, not peer reviewed; opened 2026-11-17 all records citing this
  20. Foster DJ, et al.. Advanced siRNA designs further improve in vivo performance of GalNAc-siRNA conjugates. Mol Ther (2018). paper; opened 2026-11-17 all records citing this
  21. Suntharalingam G, et al.. Cytokine storm in a phase 1 trial of the anti-CD28 monoclonal antibody TGN1412. N Engl J Med 355:1018-1028 (2006). paper; opened 2026-11-17 all records citing this
  22. Eastwood D, Findlay L, Poole S, et al.. Monoclonal antibody TGN1412 trial failure explained by species differences in CD28 expression on CD4+ effector memory T-cells. Br J Pharmacol 161:512-526 (2010). paper; opened 2026-11-17 all records citing this
History of this record (5)

This record has been changed 5 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.

Order
9
Stage
lead-optimisation
Status
partly written
Last updated
2026-11-17
Record
content/cascade/medicinal-chemistry.json