Vast fields of biologically active chemical space

Exactly 17 years ago, I highlighted a paper from Brian Shoichet’s lab in which he hypothesized that only tiny regions of chemical space are biologically relevant. If so, Practical Fragments quoted, “a major reason why the screening of synthetic compounds ever finds notable hits is that our libraries are biased toward the sort of molecules that proteins have evolved to recognize.” The best scientists continually question their assumptions, and this is exactly what Brian, Bryan Roth, and collaborators at University of California San Francisco and University of North Carolina at Chapel Hill have done in a new open-access J. Med. Chem. paper.

 

The idea that molecules more closely resembling metabolites and natural products would be more likely to be biologically active is reasonable given the promiscuity of many naturally occurring molecules. As the researchers point out, dopamine signals through 5 receptors, while serotonin binds to 14. Nature is the ultimate recycler, constantly reusing chemical motifs.

 

But since 2009 there has been a massive growth of make-on-demand libraries, and these have increasingly diverged from known molecules. Brian has been among the most prolific explorers of this new chemical space, and as he noted in his keynote at DDC 2026, these gargantuan libraries are yielding more hits against more targets.

 

Are these hits more selective? In other words, are molecules that are less “bio-like” likely to bind to fewer targets? This is the question the new paper addresses, focusing on the 5-HT_2A serotonin receptor (5-HT_2AR), the target for a variety of drugs including psychedelics. The researchers have extensively studied this GPCR; we wrote about a successful computational screen here. In the new paper, the researchers compared hits from a previous “make-on-demand” (MoD) library screen of 1.6 billion molecules with a new screen of 3.5 million “in-stock” molecules. As predicted, the in-stock compounds were much more bio-like than the MoD compounds as assessed by several metrics (Tanimoto similarity, Avalon fingerprints, and RDK7 fingerprints, for the cheminformatics aficionados among you).

 

Both libraries were computationally screened using DOCK3.8 against 5-HT_2AR. Of the top scoring hits, 85 molecules from the in-stock library and 384 molecules from the MoD library were tested in a radio-ligand displacement assay. Hit rates were nearly identical at around 24% each. Functional assays revealed the hits to be active, with some acting as agonists while others were antagonists.

 

When tested against related GPCRs, specifically 5-HT_2BR and 5-HT_2CR, the in-stock molecules were more promiscuous than the MoD molecules. But when tested against a panel of 318 GPCRs there were no differences: the in-stock molecules bound on average 1.7% of the receptors while the MoD molecules bound on average 2.1%. The similar promiscuity persisted even when focusing on the 55 aminergic GPCRs (5-HT_2AR is an aminergic GPCR). In fact, the most promiscuous compound of all came from the MoD set and is distinctly not bio-like.

 

Of course, this is just one study, but I suspect it will generalize. In 2009 I questioned whether biologically relevant chemical space is really so sparse. As I wrote then, “consider a vast field of some crop that can only be harvested by night. There are lights scattered haphazardly throughout the field. One might expect that the crops immediately under the lamp posts would be harvested more intensively than crops in darker parts of the field, even if other areas are equally productive. In this scenario, the lamp posts reveal natural products and similar molecules, but much – or even most – of (unlit) chemical space may also be biologically active, it just hasn’t been sampled yet.”

 

There is grandeur in this view of chemical space, which is supported by the new paper. Imagine yourself in the midst of the field on a moonless night. Turn on your headlamp and step forward in any direction. Drug leads glimmer as far as your light can reach.

Not fragments versus DOS, fragments from DOS

A few months ago we highlighted a forum in Nature comparing fragment-based lead discovery with diversity-oriented synthesis, or DOS. This was quite a vigorous debate and was covered on our sister blog as well as In the Pipeline and second messenger. Personally I’ve never been a this or that kind of guy – more of a this and that – so it is refreshing to see a paper in this week’s issue of PNAS describing a DOS approach to building fragments.

Damian Young and colleagues at Harvard and the Broad Institute, ground zero for DOS, noted that many commercial fragments contain a sizable percentage of sp2 carbons: aromatic rings, for example. Because a larger number of aromatic rings correlates with lower solubility and higher attrition in lead development, the researchers focused on using DOS to generate fragments that would have a higher fraction of sp3 carbons at the expense of sp2 carbons. They used a “build/couple/pair” approach, in which chiral “building blocks” (in this case proline derivatives) were “coupled” to another building block and then functional groups were “paired” to generate bicyclic molecules. The result was about three dozen fragments.

So how do they look? Actually, not so bad. Superficially they all resemble one another, but because they contain up to three stereocenters they cover quite a bit of chemical space while still conforming to the rule of 3. Significantly, they are in fact more three-dimensional than commercially available fragments (from ZINC) having the same molecular formula or the same set of calculated physical properties (molecular weight, cLogP, number of hydrogen-bond donors and acceptors, etc.). The DOS fragments contain a larger fraction of methyl esters and carboxylic acids than I would want to see in a library overall, but this was intentional, and none of them are downright ugly.

Unfortunately the paper provides no screening data, so it is anyone’s guess whether any of the fragments will turn out to be active. Still, the approach is likely to probe new areas of chemical space. Hopefully some of the commercial purveyors of fragments will start making and selling these types of molecules.

Fragments of the future - part 2 (The intersection of chemical and biological space)

In a previous post about heterocycles that appear chemically feasible but have not been reported, we wondered whether these molecules would show biological activity. The structure of biologically relevant chemical space – that fraction of possible molecules that will exhibit some biological effect – is of great interest, but as yet unknown. Brian Shoichet and coworkers at UCSF have just published a thought-provoking analysis in Nature Chemical Biology that is also relevant to developing new fragment libraries. 

The researchers ask why it is that HTS collections of a million or so compounds, vanishingly small in comparison to the roughly 1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 possible small drug-sized molecules, nonetheless so often succeed in identifying hits. A lovely paper by Tobias Fink and Jean-Louis Reymond had previously computationally enumerated all possible compounds with up to 11 C, N, O, and F atoms. Of these 26,429,328 molecules, 25,810 are commercially available. 

Shoichet and colleagues compared the structures of these compounds with the structures of metabolites and natural products (all of which have by definition been processed by at least one protein) and found that the commercially available compounds were much more similar to natural products and metabolites than were non-commercially available compounds. Indeed, the more similar a molecule is to a known natural product or metabolite, the more likely that it is available for purchase; 2918 of the commercially available compounds are in fact natural products or metabolites. 

The bias also increases exponentially with molecular size: a random 11-atom commercial compound is almost 1000-times more likely to resemble a natural product or metabolite than is a non-commercially available molecule, whereas the bias is only about 2-fold for 6-atom molecules. Similar results were observed with other libraries. 

The authors conclude that: A major reason why the screening of synthetic compounds ever finds notable hits is that our libraries are biased toward the sort of molecules that proteins have evolved to recognize.  

This resemblance is reasonable. After all, most commercially available compounds are ultimately derived from naturally occurring starting materials, so their similarity to natural products isn’t surprising. Moreover, historically much of chemistry was devoted to natural product synthesis, so many of the intermediates built up over the years resemble natural products. And of course, once you learn how to do chemistry on one moiety, you will tend to stick with it unless you have a good reason to do otherwise; each heterocycle behaves (often frustratingly) differently, so if a natural-product-like molecule does the job, why look for trouble? 

But does this “biogenic bias” mean that the rest of chemical space is a biological desert? Not necessarily. I can imagine at least two alternative models of chemical-biological diversity space. 

Let’s call one model “lamp posts in dark fields.” Consider a vast field of some crop that can only be harvested by night. There are lights scattered haphazardly throughout the field. One might expect that the crops immediately under the lamp posts would be harvested more intensively than crops in darker parts of the field, even if other areas are equally productive. In this scenario, the lamp posts reveal natural products and similar molecules, but much – or even most – of (unlit) chemical space may also be biologically active, it just hasn’t been sampled yet. 

Another possibility is the “oil-field model.” As we are all too aware, petroleum is distributed very unevenly across the globe. In some areas, such as Texas, oil was easy to find and easy to extract. In others, such as the deep ocean or the high arctic, oil is harder to find and more technically demanding to access. In this scenario, there are vast pockets of chemical space that are relevant to biology, they just haven’t been identified (let alone accessed) yet. 

These are fun, speculative questions, but the paper provides some practical data. Specifically, 83% of core ring scaffolds found in natural products are absent from commercial libraries. In fragment or lead-sized molecules of MW < 350 with less than three stereocenters, 1891 rings scaffolds found in natural products are not commercially available. These could be useful additions to fragment libraries, and the paper lists 18 examples. 

In fact, at least one company, deCODE, is explicitly enriching its fragment collection with molecules based on natural products and metabolites. This might be a good strategy. After all, even if the “lamp posts in dark fields” model is correct, there are plenty of brightly illuminated, unharvested chemotypes. At least for now, picking these may be more productive than venturing into the twilight-zone of uncharted chemical space.

More on DOCKing fragments and sampling chemical space

A few weeks ago, we highlighted a paper from Brian Shoichet’s group at UCSF demonstrating that computational screening could successfully identify fragments binding to a protein target, and that the binding modes predicted were actually observed experimentally. A companion paper just published online in PNAS now extends these results, and also beautifully illustrates that it is possible to cover much more chemical space with fragments than with lead-like molecules.

Denise Teotico, Shoichet, and colleagues used the program DOCK 3.5.54 to screen 137,639 fragments against AmpC beta-lactamase, a bacterial protein responsible for antibiotic resistance. The protein had previously been the target of HTS and computational screens of drug-like like molecules. The computational screens had modest success rates (2-7%), but the HTS screen was a total bust: of the more than 1200 hits from the 70,000+ compound screening collection, more than 95% of these turned out to be false positives, mostly aggregators, with just a few dozen true inhibitors, all of which turned out to be covalent (irreversible).

In contrast, of the 48 high-scoring fragments that were experimentally tested, 23 had Ki values better than 10 mM, for a hit rate of 48%. The authors also assessed potential for false negatives by choosing 20 random fragments and testing these for inhibition; only one showed inhibition (with a Ki value of 3.1 mM), and this molecule had scored in the top 5% of docked fragments.

The paper presents a fascinating empirical test of the Hannian chemical complexity hypothesis. Starting with the 23 active fragments, the researchers calculated how many lead-like molecules (up to 25 non-hydrogen atoms) could contain these fragments. Of the roughly 47,000,000,000 to 430,000,000,000 possible lead-like molecules, only 675 are commercially available. By repeating this analysis with fragment-sized molecules (up to 17 non-hydrogen atoms), the size of the haystack was reduced by six orders of magnitude: only about 10,000 possible molecules contain these fragments, of which 93 are commercially available. Moreover, many of the active fragments represent unique chemotypes not previously observed in AmpC inhibitors. As the authors note:

The chances of discovering interesting chemotypes for biological targets is many orders of magnitude higher when targeting molecules in the fragment weight range than even at slightly higher size ranges.

But, as the paper asks, “are the docking predictions right for the right reasons?” The researchers solved the crystal structures of 8 fragments bound to AmpC. Four of these reproduced the docking predictions well, two were somewhat different, and two were way off. In these last two cases, the protein itself adopted different conformations than had been used in the docking studies.

Protein conformational flexibility is remarkably common, and likely to be a persistent difficulty for computational methods. Clearly, current computational methods can’t identify all possibilities, particularly with fluxional proteins. Still, especially with relatively rigid proteins, computational fragment-screening may reveal chemotypes that HTS won’t.

A notable feature of the fragments is their relatively poor ligand efficiency: with one unusual exception (a phosphinate), all of the active fragments have ligand efficiencies less than 0.3 (kcal/mol)/atom. AmpC has a large, open active site, and the authors suggest that the failure of other hit-ID methods against this target may reflect issues such as solubility.

It remains to be seen whether these fragments can be advanced to low nanomolar inhibitors, but at least fragment-screening has provided many new starting points. And the paper demonstrates, once again, that triaging a fragment set computationally can be an effective means for concentrating the needles in a haystack.

Fragments in silico, confirmed by X-ray

I’ve always been something of an empiricist, and have therefore been wary of computational fragment screening. It’s not that I think it’s impossible, just that the algorithms and parameters developed to date have not often shown themselves up to the task. A paper just published in Nature Chemical Biology from Brian Shoichet’s group at UCSF has caused me to reconsider my skepticism.

Shoichet and Yu Chen used the program DOCK to screen 67,489 commercially available fragment-sized molecules contained in the database ZINC against the active site of the beta lactamase CTX-M, a bacterial enzyme responsible for resistance to penicillin and cephalosporin. Of 69 top hits, 10 actually inhibited the enzyme when tested experimentally. In contrast, of 37 high-scoring hits from a similar computational screen of 1,147,326 larger lead-like molecules, none showed any inhibition up to the limit of their solubilities.

Interestingly, each of the ten active fragments contained an anionic group: 3 carboxylates, 2 sulfates, and 5 tetrazoles among the set. A reexamination of the docked lead-like molecules revealed a relatively high-scoring tetrazole, which exhibited an experimental Ki value of 21 micromolar (see figure). Although this was an in silico hit, it was swamped by the number of (inactive) hits and so had not been selected for experimental follow-up until the fragment results revealed tetrazoles to be privileged pharmacophores. Additional similarity searching of the lead-like molecules led to two additional low micromolar inhibitors.



Five of the inhibitory fragments and one of the lead-like molecules were characterized crystallographically, and the results were remarkable: all of them bound in a similar manner to that predicted by docking.

Chen and Shoichet also investigated the specificity of the fragments compared to the lead-like compounds, and the results agreed well with those predicted by Hann and colleagues (as discussed on our sister blog FBDD-Lit here). Namely, while the fragments had relatively low specificity against a mechanistically distinct beta lactamase (AmpC), the lead-like molecule exhibited roughly 100-fold tighter inhibition of CTX-M. In other words, fragments likely have a higher hit rate (and correspondingly lower specificity) due in part to their simplicity, but as fragments are elaborated, specificity can be readily built into the molecules.

So does this mean the era of computational fragment-based screening has arrived? While these results are impressive, it is important to keep them in perspective. CTX-M has a relatively rigid active site, while many proteins of interest show a level of flexibility that confounds modeling. Moreover, Chen and Shoichet were working with an ultra-high resolution (0.88-Angstrom) crystal structure of CTX-M in which they could actually see density for hydrogen atoms on some polar groups. Needless to say, this is atypical. Still, the paper does give hope that the computational tools are ready, as long as they are applied to appropriate systems.