Commentary Special Focus: Horizons in Medicinal Chemistry For reprint orders, please contact [email protected]
To screen or not to screen: an impassioned plea for smarter chemical libraries to improve drug lead finding “It is time to perceive, from the failures of prior chemical synthetic library paradigms, a transformative path for drug discovery based on efficient, realistic and intelligent screening.” Keywords: biogenic compounds n chemical informatics n chemotypes n combinatorial chemical libraries n fragment screening n high-throughput screening n hype cycle n molecular diversity n natural products n pharmaceutical productivity
Life expectancies have been increasing in recent decades, thanks to improved medicines [1]. The pharmacopoeias, derived from natural plant extracts, go back as far as the history can trace, thanks to the Traditional Chinese Medicine and the Indian Ayurveda, among others. Despite millennia of experience, many old diseases still remain incurable, and new phenotypes are continually evolving. We still must dedicate ourselves to innovations that mitigate the resulting human and economic costs. The low-hanging fruits have been picked, but advances in cellular and molecular biology and the advent of high-throughput technologies related to drug discovery are extending the reach of the human mind in finding cures not only for the rare and neglected diseases, but also the hard to cure ailments that have been afflicting humans from time immemorial. Drug discovery over the last few decades has become a highly complex scientific endeavor, transcended from natural plant extracts to small molecules (albeit guided by the molecules found in nature), phenotypic observations to targetbased approaches, and from random, hit or miss, folklore observations to highly systematic, targeted efforts. All aspects of drug discovery have become highly centralized and industrialized with the pharmaceutical industry leading the way. The human genome has been sequenced and the NIH Roadmap Initiative was launched, while major advances in synthetic organic chemistry have empowered the combinatorial chemists to form a niche in constructing vast, empirical, combinatorially derived chemical libraries and on-demand custom libraries. Liquid handling robotics, innovative reagent and signal detection platforms, and huge improvements in bio- and chemo-informatic resources, engineered
cell lines, and so on, have all worked in concert to dramatically transform how contemporary drug-discovery research is conducted. The process of drug discovery has benefited over the last 20 years from the advent and rise of high-throughput screening (HTS), an early and vital step of screening small-molecule libraries primarily through biochemical and cell-based assays. While the discovery of new therapeutic targets in the post-human-genome-sequencing era is at an all-time high, the introduction of new molecular entities against therapeutic targets has shrunk to the lowest level in decades [2]. HTS gained popularity and prominence as a means to that end. Central to every HTS endeavor is the compound collection. Industry-style probe discovery has now gained unparalleled momentum in academia with the availability of vendorsupplied chemical libraries and ready access to institutional HTS laboratories [3]. These library collections are designed and selected for druglike properties and structural diversity, which are critical to identifying unique hits for screening targets. Millions of compounds are now commercially available. Both academia and the pharmaceutical industry have been involved in screening large compound libraries, corporate databases, virtual compound collections and suppliers’ databases for lead drug candidates. With fervor, often seemingly quite disproportionate to the apparent benefits, the pharma ceutical industry and academia have collectively screened several millions of compounds against key therapeutic targets associated with specific diseases. This cult attitude has resulted in the identification of more than 100,000 distinct chemical scaffolds interacting with more than 4000 protein targets [4]. While we attributed the dwindling pharmaceutical product pipeline to
10.4155/FMC.14.21 © 2014 Future Science Ltd
Future Med. Chem. (2014) 6(5), 497–502
Gerry Lushington1 & Rathnam Chaguturu*2 LiS Consulting, Lawrence, KS, USA iDDPartners, Princeton Junction, NJ, USA *Author for correspondence: [email protected] 1 2
ISSN 1756-8919
497
Commentary | Lushington & Chaguturu Phase II failures, an in-depth analysis points the finger at the wrong targets pursued by the pharmaceutical industry. Furthermore, the decline in pharmaceutical R&D productivity is partly attributable to its higher risk profile in recent years, focusing on rare and neglected, unmet diseases, and therapeutic targets whose disease relevance has not fully been explored or established with greater rigor. R&D efforts, in general, are on the decline due to decreased government funding, tightened budgets in industry, the inevitable patent cliff, pharmaceutical lay-offs, all of which are conspiring toward an innovation crisis that is spiraling downward uncontrollably. This has led to the industrialization of drug discovery in academia. Academic screening cores almost always rely heavily on commercially available chemical libraries. These screening sets are composed of highly generic compounds (those with publically known synthesis protocols and no compositional intellectual property value) and contain no ‘crown jewels’ (that is, novel, interesting and efficacious compounds). This is a huge barrier for the pharmaceutical industry to show any type of interest in pursuing these molecules as medicines. Deficient paradigms The 11 million-plus compounds that comprise the vendor-supplied chemical libraries are derivatives, in general, of natural product scaffolds and are designed with a biogenic bias. Even though most of the drugs that are on the market, including many of the blockbusters, have natural product origins, almost 80% of the core ring scaffolds present among the natural products are surprisingly absent in the commercially available molecules, and by extension, the screening libraries. While statistically defined chemical space is similar for natural products and marketed drugs, the combinatorially derived chemical libraries do not share this space. Compared with the natural products, combinatorially derived chemical libraries are highly deficient in chiral centers, low on oxygen atoms and rich in nitrogen – the key features that dictate target selectivity, specificity and in vivo metabolism [5,6]. The wealth of chemical diversity that has evolved with biological diversity is underrepresented in the commercial chemical library offerings, but needs to be harnessed in earnest to ease the current drug-discovery bottleneck. Since the chemical diversity of these libraries is not always relevant to biological function [7], an earnest plea is being made to the chemical library 498
Future Med. Chem. (2014) 6(5)
vendors to advance the chemical methodology and library development technology platform to increase the natural product-like attributes and thus play their part in improving the success of our lead finding efforts Natural product-based therapeutics of the yesteryear in contrast to the evidence-based and US FDA-approved medicines, was typically semipurified and enriched extracts, and not single compounds. This brings to light the distinction of ‘solo versus concert’ performance. The so-called ‘concert performance’ of natural products is more ideally suited to modulating network pharmacology from a systems biology perspective, rather than pursuing a simplistic, contemporary ‘single target–single disease’ modality. The strategy of the erstwhile CombinatoRx (now Zalicus; MA, USA), identifying novel product candidates based on unexpected biological synergies, and its belief – that multiple drugs interacting with multiple targets affect the complex biology of human disease – may be of significant value in cultivating an ‘out-of-the-box’ mindset to accelerate and expedite drug discovery, at least at the front-end [8]. ‘Drug repurposing’ and ‘phenotypic screening’ are in vogue, especially in academic drugdiscovery circles. Scientists do not get ‘the drug’ straight from a screening library. The compound identified from a screening campaign needs to be optimized, often requiring 1000- to 10,000fold enhancement in both activity and relevant pharmacokinetic (PK) and pharmacodynamic (PD) properties. Phenotypic-based protocols supplement the medicinal chemist’s need for precision in pharmacological tuning the intrinsic potency of a compound’s performance. Repurposing of failed drugs or the drugs on the market is a potentially lucrative strategy for accelerating product development, especially in terms of overcoming the safety concerns, but intellectual property imposes key disincentives for their commercialization. Intelligent library formulation If academic screens lack so-called ‘crown jewels’, while drug repurposing and phenotypic screening are desperate efforts by the old guard, where do we go from here? The answer, perhaps, is that we must mine both target and therapeutic spaces in the most exhaustive yet efficient way possible, through carefully designed screening campaigns. As a de novo discovery tool, compound screening can open broad new vistas, fence us in or send us off chasing false or non-optimizable leads: it all future science group
An impassioned plea for smarter chemical libraries to improve drug lead finding depends on the compounds we choose to look at. Throughout two decades of impressive advancement in high-throughput assay technology, the number of new therapeutic entities approved by the FDA per year has remained approximately the same as in prior decades. How could this be if we consider that compound screening efficiency has increased by a factor of thousands over the same time frame? Quite simple: compare a quarter of a million data points derived from the most sophisticated ultra-HTS platform against 25 compounds assayed by hand. If every data point from the technologically superior study is either inactive or discrepant then our project is dead. If the manual study produces high scoring hits then we’re off to the races! Technological advances are certainly not unhelpful, but they require commensurate adaptations in strategy and perspective. It is obvious that the way to exploit greater compound screening efficiency is by having more compounds to screen, but the slower and more painful lesson we have been learning is that these new compounds must be both biologically interesting and chemically diverse. If they are not biologically interesting, they will not hit targets in ways that will lead productively to therapeutics. If they do not span different chemistries, then our large-scale automated screen may tell us little that we couldn’t have deduced from a small manual one. Thus, every time we see reports of new enhancements in assay resolution and efficiency, we should ask ourselves how we are going to improve and expand our screening sets to avail ourselves of the incumbent analytical advantages. To this end, let us start with the wild assumption that chemists can synthesize and isolate anything that we ask. How might we phrase our wish list? Here are some suggestions: n New analogs of old drugs, with core and substituent permutations prioritized according to promising absorption, distribution, metabolism and excretion (ADME), toxicology and PK profiles; A product matrix of common but structurally diverse organic scaffolds according to the following specifications: each scaffold must have demonstrably interesting activities in at least one assay (but not be active for more than several distinct targets) and be decorated with a standardized set of substituents chosen to span a reasonable spectrum of chemical functionality while retaining decent solubility and cell permeability profiles;
n
future science group
| Commentary
A collection of diverse heterocycles, including ones whose scaffolds are drawn from known privileged structures, are reminiscent of privileged structures, and perhaps including some with little track record in prior screening as long as there are no previously characterized alert flags indicative of toxicity, compound aggregation, undue chemical reactivity or serious ADME liabilities;
n
A fragment screening set of small heterocycles, small undecorated polycyclics, plus various common organic linkers and branching groups that (at the minimum) collectively embrace the substructural functionality evident in known small-molecule drugs;
n
A collection of small linear and cyclic peptides, including structures with previously established bioactivity, plus novel oligomers based on helix, strand, turn and coil structures associated with known protein-interaction interfaces;
n
An extensive and diverse collection of isolated natural products embracing both known actives as well as novel compounds with promising solubility and cell permeability profiles;
n
A battery of constitutionally characterized and enriched natural product extracts.
n
The bulk of many compound screening sets, and a substantial fraction of the above idealized library, focuses on conventional organic chemical synthesis products, primarily because such species typically have a better track record than peptides in producing viable therapeutics, and are generally easier to synthesize and isolate than most natural products. The drawback with such compounds is that the manifold of synthetically tractable compounds is extraordinarily large, perhaps on the order of 1060 distinct compounds [9]. It is impossible to systematically profile such a manifold, even via computational means. This can be overcome, however, by examining the behavior of fragments. Fragment screening assumes, based on fair empirical evidence, that total ligand-binding affinity exhibits a somewhat additive relationship with the affinities of constituent chemical subgroups. Thus, the large body of 1060 distinct compounds can be viewed as a permutative assembly of a much smaller number of distinct fragments, and the bioactivity of each of the permutative products is roughly related to the bioactivities of component www.future-science.com
499
Commentary | Lushington & Chaguturu fragments. Thus, a preliminary fragment screen on a given target should provide a useful basis for reducing the immense body of potentially active compounds down to a much smaller subset whose activity prospects are statistically greatly enhanced. Resources & strategies for library refinement The aforementioned wish list of desirable screening compounds may sound rational and comprehensive, but the precise formulation of a real library rests on assumptions that may not yet be practicable. How can we assume that a compound that has not yet been thoroughly characterized in vivo will have suitable ADME, toxicology and PK profiles? How exactly do we ensure structural diversity? How can we infer from known privileged heterocycles other scaffolds that will have similar biological attributes? Fortunately, the ambitious data collection associated with the NIH Roadmap has generated a strong basis for answering many of these questions through statistical inference and data mining. At the beginning of the year, the PubChem data repository had collected information on more than 744,000 separate assay experiments, while storing information on nearly 100 times as many distinct chemical entities (72,575,000). These millions of unique bioactivity data points collectively provide a strong basis for evaluating tested compounds according to their prospects for effecting interesting behavior in assays that those compounds have not yet been subjected to. By statistical arguments alone, one may apply the bioactivity data compiled for compounds already in PubChem to develop a screening set comprised of compounds with some optimal degree of proven biological relevance (i.e., at least some bioactivity, but not promiscuous). A zeroth-order approximation to such a screening set could be the compound collection supplied for assays supported by the Molecular Libraries Initiative (MLI; the government program that originally commissioned PubChem): this MLI set has the advantage of having been exhaustively tested, thus providing the informational basis for identifying serially inactive compounds and frequent-hitters. It is unlikely, however, that the set would afford sufficient coverage of all of the potentially desirable chemical families that we have identified. While the MLI set possesses a greater biogenic bias than is evident in comparably sized commercial compound sets, it is generally deficient in peptidic and peptidomimetic substances, underrepresents 500
Future Med. Chem. (2014) 6(5)
various classes of natural products and was never systematically optimized for chemical diversity [10]. By way of perspective, when the NCI formulated its NExT diversity library [11] based on compounds from the original MLI screening set and various other sources, only about 12,000 MLI compounds (
To screen or not to screen: an impassioned plea for smarter chemical libraries to improve drug lead finding “It is time to perceive, from the failures of prior chemical synthetic library paradigms, a transformative path for drug discovery based on efficient, realistic and intelligent screening.” Keywords: biogenic compounds n chemical informatics n chemotypes n combinatorial chemical libraries n fragment screening n high-throughput screening n hype cycle n molecular diversity n natural products n pharmaceutical productivity
Life expectancies have been increasing in recent decades, thanks to improved medicines [1]. The pharmacopoeias, derived from natural plant extracts, go back as far as the history can trace, thanks to the Traditional Chinese Medicine and the Indian Ayurveda, among others. Despite millennia of experience, many old diseases still remain incurable, and new phenotypes are continually evolving. We still must dedicate ourselves to innovations that mitigate the resulting human and economic costs. The low-hanging fruits have been picked, but advances in cellular and molecular biology and the advent of high-throughput technologies related to drug discovery are extending the reach of the human mind in finding cures not only for the rare and neglected diseases, but also the hard to cure ailments that have been afflicting humans from time immemorial. Drug discovery over the last few decades has become a highly complex scientific endeavor, transcended from natural plant extracts to small molecules (albeit guided by the molecules found in nature), phenotypic observations to targetbased approaches, and from random, hit or miss, folklore observations to highly systematic, targeted efforts. All aspects of drug discovery have become highly centralized and industrialized with the pharmaceutical industry leading the way. The human genome has been sequenced and the NIH Roadmap Initiative was launched, while major advances in synthetic organic chemistry have empowered the combinatorial chemists to form a niche in constructing vast, empirical, combinatorially derived chemical libraries and on-demand custom libraries. Liquid handling robotics, innovative reagent and signal detection platforms, and huge improvements in bio- and chemo-informatic resources, engineered
cell lines, and so on, have all worked in concert to dramatically transform how contemporary drug-discovery research is conducted. The process of drug discovery has benefited over the last 20 years from the advent and rise of high-throughput screening (HTS), an early and vital step of screening small-molecule libraries primarily through biochemical and cell-based assays. While the discovery of new therapeutic targets in the post-human-genome-sequencing era is at an all-time high, the introduction of new molecular entities against therapeutic targets has shrunk to the lowest level in decades [2]. HTS gained popularity and prominence as a means to that end. Central to every HTS endeavor is the compound collection. Industry-style probe discovery has now gained unparalleled momentum in academia with the availability of vendorsupplied chemical libraries and ready access to institutional HTS laboratories [3]. These library collections are designed and selected for druglike properties and structural diversity, which are critical to identifying unique hits for screening targets. Millions of compounds are now commercially available. Both academia and the pharmaceutical industry have been involved in screening large compound libraries, corporate databases, virtual compound collections and suppliers’ databases for lead drug candidates. With fervor, often seemingly quite disproportionate to the apparent benefits, the pharma ceutical industry and academia have collectively screened several millions of compounds against key therapeutic targets associated with specific diseases. This cult attitude has resulted in the identification of more than 100,000 distinct chemical scaffolds interacting with more than 4000 protein targets [4]. While we attributed the dwindling pharmaceutical product pipeline to
10.4155/FMC.14.21 © 2014 Future Science Ltd
Future Med. Chem. (2014) 6(5), 497–502
Gerry Lushington1 & Rathnam Chaguturu*2 LiS Consulting, Lawrence, KS, USA iDDPartners, Princeton Junction, NJ, USA *Author for correspondence: [email protected] 1 2
ISSN 1756-8919
497
Commentary | Lushington & Chaguturu Phase II failures, an in-depth analysis points the finger at the wrong targets pursued by the pharmaceutical industry. Furthermore, the decline in pharmaceutical R&D productivity is partly attributable to its higher risk profile in recent years, focusing on rare and neglected, unmet diseases, and therapeutic targets whose disease relevance has not fully been explored or established with greater rigor. R&D efforts, in general, are on the decline due to decreased government funding, tightened budgets in industry, the inevitable patent cliff, pharmaceutical lay-offs, all of which are conspiring toward an innovation crisis that is spiraling downward uncontrollably. This has led to the industrialization of drug discovery in academia. Academic screening cores almost always rely heavily on commercially available chemical libraries. These screening sets are composed of highly generic compounds (those with publically known synthesis protocols and no compositional intellectual property value) and contain no ‘crown jewels’ (that is, novel, interesting and efficacious compounds). This is a huge barrier for the pharmaceutical industry to show any type of interest in pursuing these molecules as medicines. Deficient paradigms The 11 million-plus compounds that comprise the vendor-supplied chemical libraries are derivatives, in general, of natural product scaffolds and are designed with a biogenic bias. Even though most of the drugs that are on the market, including many of the blockbusters, have natural product origins, almost 80% of the core ring scaffolds present among the natural products are surprisingly absent in the commercially available molecules, and by extension, the screening libraries. While statistically defined chemical space is similar for natural products and marketed drugs, the combinatorially derived chemical libraries do not share this space. Compared with the natural products, combinatorially derived chemical libraries are highly deficient in chiral centers, low on oxygen atoms and rich in nitrogen – the key features that dictate target selectivity, specificity and in vivo metabolism [5,6]. The wealth of chemical diversity that has evolved with biological diversity is underrepresented in the commercial chemical library offerings, but needs to be harnessed in earnest to ease the current drug-discovery bottleneck. Since the chemical diversity of these libraries is not always relevant to biological function [7], an earnest plea is being made to the chemical library 498
Future Med. Chem. (2014) 6(5)
vendors to advance the chemical methodology and library development technology platform to increase the natural product-like attributes and thus play their part in improving the success of our lead finding efforts Natural product-based therapeutics of the yesteryear in contrast to the evidence-based and US FDA-approved medicines, was typically semipurified and enriched extracts, and not single compounds. This brings to light the distinction of ‘solo versus concert’ performance. The so-called ‘concert performance’ of natural products is more ideally suited to modulating network pharmacology from a systems biology perspective, rather than pursuing a simplistic, contemporary ‘single target–single disease’ modality. The strategy of the erstwhile CombinatoRx (now Zalicus; MA, USA), identifying novel product candidates based on unexpected biological synergies, and its belief – that multiple drugs interacting with multiple targets affect the complex biology of human disease – may be of significant value in cultivating an ‘out-of-the-box’ mindset to accelerate and expedite drug discovery, at least at the front-end [8]. ‘Drug repurposing’ and ‘phenotypic screening’ are in vogue, especially in academic drugdiscovery circles. Scientists do not get ‘the drug’ straight from a screening library. The compound identified from a screening campaign needs to be optimized, often requiring 1000- to 10,000fold enhancement in both activity and relevant pharmacokinetic (PK) and pharmacodynamic (PD) properties. Phenotypic-based protocols supplement the medicinal chemist’s need for precision in pharmacological tuning the intrinsic potency of a compound’s performance. Repurposing of failed drugs or the drugs on the market is a potentially lucrative strategy for accelerating product development, especially in terms of overcoming the safety concerns, but intellectual property imposes key disincentives for their commercialization. Intelligent library formulation If academic screens lack so-called ‘crown jewels’, while drug repurposing and phenotypic screening are desperate efforts by the old guard, where do we go from here? The answer, perhaps, is that we must mine both target and therapeutic spaces in the most exhaustive yet efficient way possible, through carefully designed screening campaigns. As a de novo discovery tool, compound screening can open broad new vistas, fence us in or send us off chasing false or non-optimizable leads: it all future science group
An impassioned plea for smarter chemical libraries to improve drug lead finding depends on the compounds we choose to look at. Throughout two decades of impressive advancement in high-throughput assay technology, the number of new therapeutic entities approved by the FDA per year has remained approximately the same as in prior decades. How could this be if we consider that compound screening efficiency has increased by a factor of thousands over the same time frame? Quite simple: compare a quarter of a million data points derived from the most sophisticated ultra-HTS platform against 25 compounds assayed by hand. If every data point from the technologically superior study is either inactive or discrepant then our project is dead. If the manual study produces high scoring hits then we’re off to the races! Technological advances are certainly not unhelpful, but they require commensurate adaptations in strategy and perspective. It is obvious that the way to exploit greater compound screening efficiency is by having more compounds to screen, but the slower and more painful lesson we have been learning is that these new compounds must be both biologically interesting and chemically diverse. If they are not biologically interesting, they will not hit targets in ways that will lead productively to therapeutics. If they do not span different chemistries, then our large-scale automated screen may tell us little that we couldn’t have deduced from a small manual one. Thus, every time we see reports of new enhancements in assay resolution and efficiency, we should ask ourselves how we are going to improve and expand our screening sets to avail ourselves of the incumbent analytical advantages. To this end, let us start with the wild assumption that chemists can synthesize and isolate anything that we ask. How might we phrase our wish list? Here are some suggestions: n New analogs of old drugs, with core and substituent permutations prioritized according to promising absorption, distribution, metabolism and excretion (ADME), toxicology and PK profiles; A product matrix of common but structurally diverse organic scaffolds according to the following specifications: each scaffold must have demonstrably interesting activities in at least one assay (but not be active for more than several distinct targets) and be decorated with a standardized set of substituents chosen to span a reasonable spectrum of chemical functionality while retaining decent solubility and cell permeability profiles;
n
future science group
| Commentary
A collection of diverse heterocycles, including ones whose scaffolds are drawn from known privileged structures, are reminiscent of privileged structures, and perhaps including some with little track record in prior screening as long as there are no previously characterized alert flags indicative of toxicity, compound aggregation, undue chemical reactivity or serious ADME liabilities;
n
A fragment screening set of small heterocycles, small undecorated polycyclics, plus various common organic linkers and branching groups that (at the minimum) collectively embrace the substructural functionality evident in known small-molecule drugs;
n
A collection of small linear and cyclic peptides, including structures with previously established bioactivity, plus novel oligomers based on helix, strand, turn and coil structures associated with known protein-interaction interfaces;
n
An extensive and diverse collection of isolated natural products embracing both known actives as well as novel compounds with promising solubility and cell permeability profiles;
n
A battery of constitutionally characterized and enriched natural product extracts.
n
The bulk of many compound screening sets, and a substantial fraction of the above idealized library, focuses on conventional organic chemical synthesis products, primarily because such species typically have a better track record than peptides in producing viable therapeutics, and are generally easier to synthesize and isolate than most natural products. The drawback with such compounds is that the manifold of synthetically tractable compounds is extraordinarily large, perhaps on the order of 1060 distinct compounds [9]. It is impossible to systematically profile such a manifold, even via computational means. This can be overcome, however, by examining the behavior of fragments. Fragment screening assumes, based on fair empirical evidence, that total ligand-binding affinity exhibits a somewhat additive relationship with the affinities of constituent chemical subgroups. Thus, the large body of 1060 distinct compounds can be viewed as a permutative assembly of a much smaller number of distinct fragments, and the bioactivity of each of the permutative products is roughly related to the bioactivities of component www.future-science.com
499
Commentary | Lushington & Chaguturu fragments. Thus, a preliminary fragment screen on a given target should provide a useful basis for reducing the immense body of potentially active compounds down to a much smaller subset whose activity prospects are statistically greatly enhanced. Resources & strategies for library refinement The aforementioned wish list of desirable screening compounds may sound rational and comprehensive, but the precise formulation of a real library rests on assumptions that may not yet be practicable. How can we assume that a compound that has not yet been thoroughly characterized in vivo will have suitable ADME, toxicology and PK profiles? How exactly do we ensure structural diversity? How can we infer from known privileged heterocycles other scaffolds that will have similar biological attributes? Fortunately, the ambitious data collection associated with the NIH Roadmap has generated a strong basis for answering many of these questions through statistical inference and data mining. At the beginning of the year, the PubChem data repository had collected information on more than 744,000 separate assay experiments, while storing information on nearly 100 times as many distinct chemical entities (72,575,000). These millions of unique bioactivity data points collectively provide a strong basis for evaluating tested compounds according to their prospects for effecting interesting behavior in assays that those compounds have not yet been subjected to. By statistical arguments alone, one may apply the bioactivity data compiled for compounds already in PubChem to develop a screening set comprised of compounds with some optimal degree of proven biological relevance (i.e., at least some bioactivity, but not promiscuous). A zeroth-order approximation to such a screening set could be the compound collection supplied for assays supported by the Molecular Libraries Initiative (MLI; the government program that originally commissioned PubChem): this MLI set has the advantage of having been exhaustively tested, thus providing the informational basis for identifying serially inactive compounds and frequent-hitters. It is unlikely, however, that the set would afford sufficient coverage of all of the potentially desirable chemical families that we have identified. While the MLI set possesses a greater biogenic bias than is evident in comparably sized commercial compound sets, it is generally deficient in peptidic and peptidomimetic substances, underrepresents 500
Future Med. Chem. (2014) 6(5)
various classes of natural products and was never systematically optimized for chemical diversity [10]. By way of perspective, when the NCI formulated its NExT diversity library [11] based on compounds from the original MLI screening set and various other sources, only about 12,000 MLI compounds (
