Regenerative
The strongest peptide binder is not always the best drug candidate
By John Hyland, . Life and Health Today.
The strongest binder in a peptide drug screen is not necessarily the best place to start building a medicine. That is the central argument of a commentary published by Genetic Engineering and Biotechnology News, written from the perspective of researchers designing peptide discovery campaigns.
Peptides are short chains of amino acids, the building blocks of proteins, that can be engineered to latch onto specific targets in the body. They sit between small-molecule drugs and large biologics like antibodies in size, and interest in them has grown as tools for building and screening large libraries have improved. But the commentary argues that the field has organised itself around a narrow goal: find the molecule that binds most tightly.
Binding affinity, the report says, is important but is not a drug profile on its own. A peptide must also be selective enough not to hit the wrong targets, stable enough to survive in the body, soluble enough to be delivered, and capable of reaching the right biological compartment, which, for targets inside a cell, means crossing a membrane. A screen that measures only binding leaves all of those questions for a later stage, when changing direction is more costly.
The commentary, attributed to Sethera Therapeutics in an accompanying figure caption, points to a 2026 study that screened 15,360 fully random cyclic peptides, ring-shaped rather than linear chains, which tend to be more stable, against an intracellular protein interaction called Keap1-Nrf2. Starting from that screen, the researchers used iterative rounds of design, synthesis and testing to produce a molecule that could cross a cell membrane and remain active inside living cells. Genetic Engineering and Biotechnology News describes the broader lesson as showing that binding and cell entry "can be treated as connected design problems rather than sequential hurdles."
The piece also challenges how screening data are recorded and used. A peptide that fails a screen, it argues, may have failed for any number of reasons: it was not synthesised efficiently, did not display correctly in the assay, aggregated, degraded, or reached its target without changing its function. Collapsing all of those outcomes into a single "inactive" label, the commentary says, produces datasets that are difficult to learn from, because a computational model trained on them cannot distinguish a technical failure from a genuine biological one.
On the computational side, the report argues that physics-based modelling and machine learning are more useful as complementary tools than as rivals. Physics-based calculations can examine how a molecule folds and contacts its target; machine learning can identify patterns across larger datasets and propose combinations a research team might not prioritise. Recent work, the commentary notes, has coupled generative models, software that proposes new molecular structures, with experimental testing cycles to improve peptide scaffolds under defined constraints, though it flags that limited training data remain a central challenge for deep-learning approaches to cyclic peptides.
What this piece does not establish is whether any of the design principles it advocates have been tested head-to-head against conventional screening approaches in a controlled study. The argument is methodological and draws on illustrative examples rather than a systematic comparison. The 2026 cyclic peptide study it cites produced a cell-active molecule in the laboratory; Genetic Engineering and Biotechnology News does not report that it has entered human trials or received regulatory review.
Any limitations or conflicts of interest the authors declared in the closing section of the commentary are not covered in this report. Readers who want the full account should consult the original publication directly.
For anyone following peptide therapeutics as a field, the practical question the piece raises is whether the industry's standard metrics for early-stage screening are selecting for the right properties. That question will only be answered by programmes that track molecules from screen to clinic and report honestly on which early signals predicted late success. None of that data is cited here.