Oscillation detection with period uncertainty and a limited sampling budget
Equally spaced temporal sampling is the standard protocol for the study of biological rhythms. These equispaced designs perform well when calibrated to an oscillators period yet can have systematic detection biases when applied to rhythms of unknown periodicity. Here, we present a broadly-applicable set of computational methods for seeking optimal measurement schedules for rhythm detection. Our PowerCHORD methods generate experimental designs by maximizing a closed-form expression for the statistical power of the cosinor model using a black-box optimization method (differential evolution), a brute-force search, or mixed-integer conic programming. Application of these three methods showed numerically that they improve upon equispaced designs under many experimental contexts. Our numerical results also revealed an intuitive approach for achieving optimal power for simultaneous investigation of circadian, circalunar, and circannual rhythms. Our findings suggest that timing optimization is an effective yet under-explored tool for improving biological rhythm discovery.