We want to add reductionist methodologies to John et al.'s valuable list of the various contexts where proxy failure can be observed. Reductionist methodologies primarily focus on a lower-level component of their phenomena of interest with this presumption that understanding or manipulating that component is sufficient for understanding or manipulating their phenomenon of interest. In other words, they presume that their phenomenon of interest can be reduced to some of its lower-level components. We believe that extending the conceptualization of John et al. to reductionist methodologies and using the insights it might offer can be of great value because currently, despite considerable criticism, reductionism dominates diverse scientific fields; from psychology and neuroscience to biomedical sciences and drug discovery.
To align these methodologies with the conceptualization and terminology of John et al., researchers can be considered regulators, with the goal of understanding or manipulating some complex phenomenon manifested by their objects of study (agent). These objects of study “possess multiple ways of producing or expressing a proxy, which can be influenced by feedback from the regulator” (target article, sect. 3.1, para. 4). Finally, the proxy is a lower-level component of the phenomenon under study that the researchers prioritize. The pressure “that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1) has been provided and amplified by the deep-seated reductionist mindset of the scientific community and the incessant technological progress that has enabled the dissection and reduction of many phenomena to their lower levels.
Consider the example of target-based drug discovery that has been the dominant paradigm of drug discovery for about four decades. Although the goal of researchers in the field of drug discovery (regulators) is to discover molecules that can suitably alter the phenotypes of humans, based on reductionist target-based drug discovery, the lower-level proxy of binding affinity to a target protein is prioritized. Instead of selecting and optimizing molecules based on their effects on phenotypes, candidate molecules are primarily selected and optimized based on their binding affinity to a target protein whose manipulation is supposed to counteract the disorder; for example, candidate antipsychotics are selected and optimized based on their binding affinity to specific dopamine receptors. This selection and optimization is the regulatory feedback mentioned by John et al. where the agents are selected based on the proxy and “which induces optimization of the proxy” (target article, sect. 3.3, para. 1).
For years, binding affinity to a target has been criticized as an overly simplistic proxy (Horrobin, Reference Horrobin2003). Our recent analysis of the real-world efficiency of target-based drug discovery further substantiates these criticisms and reveals, based on significant evidence, the failure of this proxy (Sadri, Reference Sadri2023). The data reveal that, despite decades of utter dominance, only 9.4% of small-molecule-approved drugs have originated from target-based drug discovery.
Moreover, the analysis demonstrates that even this minor portion cannot be entirely attributed to target-based drug discovery, as their therapeutic effects are mediated by numerous mechanisms that are independent of the targets they have been discovered for (Sadri, Reference Sadri2023). This aligns perfectly with one of the factors identified by John et al. to drive proxy failure: The human body and the therapeutic effects of molecules are highly complex and there are numerous “proxy-independent actions that lead to the goal” (target article, sect. 3.3, para. 1).
Another match is between the evolution of drug discovery methodologies and the statement of John et al. that “whenever incentivization or selection is based on an imperfect proxy measure of the underlying goal, a pressure arises that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1). Target-based drug discovery was born out of rational drug discovery (Al-Ali, Reference Al-Ali2016; Sadri, Reference Sadri2023). Rational drug discovery used the available scientific knowledge, from molecular biology and physiology to pathology and pharmacology, to guide and focus the random screening of substances. This approach was immensely successful and culminated in the discovery of tens of approved drugs and Nobel Prizes for several of its pioneers: Paul Ehrlich, Gertrude Elion, George Hitchings, and James Black. Assessing the activity of molecules at the protein level and their binding affinity toward specific proteins was an important proxy and source of information in rational drug discovery; however, in the end, it relied on phenotypic observations for selecting and optimizing molecules. This is evident in the discoveries of the abovementioned pioneers and many other examples (Sadri, Reference Sadri2023). Gradually, the proxy of binding affinity to targets got excessively prioritized to the point that it even replaced the use of phenotypic observations in selecting and optimizing molecules.
Similar cases of proxy failure can be recognized in other reductionist methodologies across different fields. For example, in neuroscience, although the goal is to understand the behaviors of organisms (agents), researchers (regulators) excessively prioritize various proxies and observations at lower levels such as neurons and neural circuits (Frangou, Reference Frangou2020; Gazzaniga, Reference Gazzaniga2010; Krakauer et al., Reference Krakauer, Ghazanfar, Gomez-Marin, MacIver and Poeppel2017; Parker, Reference Parker2022; Uttal, Reference Uttal2003). This conceptualization can be extended to the reductionist methodologies that are currently dominant across various fields, including molecular biology (Lazebnik, Reference Lazebnik2002) and medical sciences (Ahn et al., Reference Ahn, Tewari, Poon and Phillips2006).
However, it must be noted that although John et al.'s conceptualization perfectly fits the reductionist methodology of target-based drug discovery, extending it to other reductionist methodologies may require a more general account of proxy failure. The challenge is generalizing to these methodologies the key concept of the pressure “that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1). Alternatively, it can be proposed that another approach toward generalizing the conceptualization of John et al. to reductionist methodologies might mitigate this challenge. For example, reductionism and the reductionist methodologies themselves can be considered as the regulator, the researchers as the agents, investigating the object of study as the goal, and the lower-level observations as the proxies.
Anyhow, we believe that the concept of proxy failure is a valuable asset for recognizing and addressing the limitations of reductionism, which is in absolute reign across diverse fields. Lest these limitations are addressed, huge resources would be expended on research with marginal contact with the real world.
You have
Access
We want to add reductionist methodologies to John et al.'s valuable list of the various contexts where proxy failure can be observed. Reductionist methodologies primarily focus on a lower-level component of their phenomena of interest with this presumption that understanding or manipulating that component is sufficient for understanding or manipulating their phenomenon of interest. In other words, they presume that their phenomenon of interest can be reduced to some of its lower-level components. We believe that extending the conceptualization of John et al. to reductionist methodologies and using the insights it might offer can be of great value because currently, despite considerable criticism, reductionism dominates diverse scientific fields; from psychology and neuroscience to biomedical sciences and drug discovery.
To align these methodologies with the conceptualization and terminology of John et al., researchers can be considered regulators, with the goal of understanding or manipulating some complex phenomenon manifested by their objects of study (agent). These objects of study “possess multiple ways of producing or expressing a proxy, which can be influenced by feedback from the regulator” (target article, sect. 3.1, para. 4). Finally, the proxy is a lower-level component of the phenomenon under study that the researchers prioritize. The pressure “that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1) has been provided and amplified by the deep-seated reductionist mindset of the scientific community and the incessant technological progress that has enabled the dissection and reduction of many phenomena to their lower levels.
Consider the example of target-based drug discovery that has been the dominant paradigm of drug discovery for about four decades. Although the goal of researchers in the field of drug discovery (regulators) is to discover molecules that can suitably alter the phenotypes of humans, based on reductionist target-based drug discovery, the lower-level proxy of binding affinity to a target protein is prioritized. Instead of selecting and optimizing molecules based on their effects on phenotypes, candidate molecules are primarily selected and optimized based on their binding affinity to a target protein whose manipulation is supposed to counteract the disorder; for example, candidate antipsychotics are selected and optimized based on their binding affinity to specific dopamine receptors. This selection and optimization is the regulatory feedback mentioned by John et al. where the agents are selected based on the proxy and “which induces optimization of the proxy” (target article, sect. 3.3, para. 1).
For years, binding affinity to a target has been criticized as an overly simplistic proxy (Horrobin, Reference Horrobin2003). Our recent analysis of the real-world efficiency of target-based drug discovery further substantiates these criticisms and reveals, based on significant evidence, the failure of this proxy (Sadri, Reference Sadri2023). The data reveal that, despite decades of utter dominance, only 9.4% of small-molecule-approved drugs have originated from target-based drug discovery.
Moreover, the analysis demonstrates that even this minor portion cannot be entirely attributed to target-based drug discovery, as their therapeutic effects are mediated by numerous mechanisms that are independent of the targets they have been discovered for (Sadri, Reference Sadri2023). This aligns perfectly with one of the factors identified by John et al. to drive proxy failure: The human body and the therapeutic effects of molecules are highly complex and there are numerous “proxy-independent actions that lead to the goal” (target article, sect. 3.3, para. 1).
Another match is between the evolution of drug discovery methodologies and the statement of John et al. that “whenever incentivization or selection is based on an imperfect proxy measure of the underlying goal, a pressure arises that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1). Target-based drug discovery was born out of rational drug discovery (Al-Ali, Reference Al-Ali2016; Sadri, Reference Sadri2023). Rational drug discovery used the available scientific knowledge, from molecular biology and physiology to pathology and pharmacology, to guide and focus the random screening of substances. This approach was immensely successful and culminated in the discovery of tens of approved drugs and Nobel Prizes for several of its pioneers: Paul Ehrlich, Gertrude Elion, George Hitchings, and James Black. Assessing the activity of molecules at the protein level and their binding affinity toward specific proteins was an important proxy and source of information in rational drug discovery; however, in the end, it relied on phenotypic observations for selecting and optimizing molecules. This is evident in the discoveries of the abovementioned pioneers and many other examples (Sadri, Reference Sadri2023). Gradually, the proxy of binding affinity to targets got excessively prioritized to the point that it even replaced the use of phenotypic observations in selecting and optimizing molecules.
Similar cases of proxy failure can be recognized in other reductionist methodologies across different fields. For example, in neuroscience, although the goal is to understand the behaviors of organisms (agents), researchers (regulators) excessively prioritize various proxies and observations at lower levels such as neurons and neural circuits (Frangou, Reference Frangou2020; Gazzaniga, Reference Gazzaniga2010; Krakauer et al., Reference Krakauer, Ghazanfar, Gomez-Marin, MacIver and Poeppel2017; Parker, Reference Parker2022; Uttal, Reference Uttal2003). This conceptualization can be extended to the reductionist methodologies that are currently dominant across various fields, including molecular biology (Lazebnik, Reference Lazebnik2002) and medical sciences (Ahn et al., Reference Ahn, Tewari, Poon and Phillips2006).
However, it must be noted that although John et al.'s conceptualization perfectly fits the reductionist methodology of target-based drug discovery, extending it to other reductionist methodologies may require a more general account of proxy failure. The challenge is generalizing to these methodologies the key concept of the pressure “that tends to make the proxy a worse approximation of the goal” (target article, abstract, para. 1). Alternatively, it can be proposed that another approach toward generalizing the conceptualization of John et al. to reductionist methodologies might mitigate this challenge. For example, reductionism and the reductionist methodologies themselves can be considered as the regulator, the researchers as the agents, investigating the object of study as the goal, and the lower-level observations as the proxies.
Anyhow, we believe that the concept of proxy failure is a valuable asset for recognizing and addressing the limitations of reductionism, which is in absolute reign across diverse fields. Lest these limitations are addressed, huge resources would be expended on research with marginal contact with the real world.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.