PumpKin: A machine-learning pipeline for automatically tracking localized kinematics in freely moving C. elegans
Source: PLOS (Public Library of Science) · 2026-07-17
Author summary Studies in neuroscience often seek to connect neural representations of sensory information with a corresponding behavioral response. These measured behaviors may be divided into two groups: those that involve multiple limbs and are easy to measure (e.g., locomotion) and those that involve only a single limb or organ and are more difficult to measure (e.g., swallowing food). Artificial intelligence (AI) has enabled the detailed study of these gross and localized behaviors. However, these technologies alone can struggle to track certain behaviors that are localized to a single limb or organ as a result of the behaviors’ high sensitivity to noise. One example of a localized behavior is found in the microscopic nematode C. elegans, where a pharyngeal organ called the grinder is measured to estimate the pumping (feeding) rate of the worms. The grinder is < 30 μm wide, less than the width of the average human hair, and moves at a speed of 5 pumps per second, equivalent to five times the average human heart rate. Thus, tracking this organ requires not only a sophisticated tracker, but also a program that can remove noise from the measured position of the grinder. To address this need, we introduce PumpKin: an open-source package designed to reliably measure the pumping rate of C. elegans.
Introduction Behaviors that are localized to a single body structure are critical to our understanding of larger behavioral patterns and decision-making. For example, the timing of blinking, although often perceived as spontaneous, is crucial for mai... [37129 chars]