More than meets the AI: The possibilities and limits of machine learning in olfaction

Abstract

Can machine learning crack the code in the nose? Over the past decade, studies tried to solve the relation between chemical structure and sensory quality with Big Data. These studies advanced computational models of the olfactory stimulus, utilizing artificial intelligence to mine for clear correlations between chemistry and psychophysics. Computational perspectives promised to solve the mystery of olfaction with more data and better data processing tools. None of them succeeded, however, and it matters as to why this is the case. This article argues that we should be deeply skeptical about the trend to black-box the sensory system's biology in our theories of perception. Instead, we need to ground both stimulus models and psychophysical data on real causal-mechanistic explanations of the olfactory system. The central question is: Would knowledge of biology lead to a better understanding of the stimulus in odor coding than the one utilized in current machine learning models? That is indeed the case. Recent studies about receptor behavior have revealed that the olfactory system operates by principles not captured in current stimulus-response models. This may require a fundamental revision of computational approaches to olfaction, including its psychological effects. To analyze the different research programs in olfaction, we draw on Lloyd's "Logic of Research Questions," a philosophical framework which assists scientists in explicating the reasoning, conceptual commitments, and problems of a modeling approach in question.

Keywords: logic of research questions; mechanisms; medicinal chemistry; neurobiology; philosophy of science; receptor modeling; stimulus response; structure odor relationship.

FIGURE 1

Illustration of the “big data” challenge in modeling olfaction [Graph, right, by Bair (2015)]. Humans feature about 400 receptor genes. Each receptor interacts combinatorically with multiple physico-chemical properties of multiple odorants (Malnic et al., 1999) [The precise number of possible stimuli detected by the olfactory system is unknown and debated (Meister, 2015); while it is considered high (Ohloff et al., 2012)]. Additionally, there are several thousand possible parameters involved in odorant-receptor binding, resulting in an explosion of possible combinations for stimulus-receptor interactions.

FIGURE 2

Historical timeline of selected key events in olfaction research (with focus on receptor biology). For details on the history of olfaction see Barwich (2020a, Ch. 1; biology-centered); and Ohloff et al. (2012, Ch. 1; chemistry-centered). For details on the history of genetics see Kay (2000) and GPCRs see Barwich and Bschir (2017). For a review of current research on odor coding mechanisms see Kurian et al. (2021).

FIGURE 3

Keller et al.’s (2017) experimental design.

FIGURE 4

Left: Structural comparison of the 3 esters (Poivet et al., 2018). Right [top: TPSA rule (Poivet et al., 2016)]; bottom: 2 examples of receptor co-activation in ketone pairs (Poivet et al., 2018).

FIGURE 5

Classification of odorant similarity according to analytic and medicinal chemistry. Illustrated is the difference between the classification of the chemical similarity of ketones (left) according to (A) the principles of analytic chemistry and (B) medicinal chemistry and receptor behavior [image from Poivet et al. (2016); OSN stands for Olfactory Sensory Neurons]. Differences are especially visible when comparing the closest similarity pairs in (A) analytic chemistry {4; 6} and (B) medicinal chemistry {1; 2}. These differences are grounded in varying ordering criteria of chemical similarity between chemistry and medicinal chemistry [example right; (C)] [Note: Sample molecules in panel (C) are not ketones but chosen merely for an illustration of the selection criteria].