Abstract

Information In Localisation Microscopy

Patrick Fox-Roberts, PostDoc, Kings College London

Super-resolution localisation microscopy allows the creation of high resolution reconstructions of biological samples, both fixed and live, using only optical wavlength light. The price it pays for this is a reliance on post-processing to make sense of the data - and if the algorithm used for this is in some way inappropriate to the dataset, the reconstruction may contain apparent structure that, despite appearing convincing, does not represent the shape of the underlying sample. Our work has focused on finding the limits of what can be achieved using localisation microscopy, and developing better algorithms that get us closer to that goal. Bayesian Blinking and Bleaching (3B) provides a fully Bayesian framework for inferring structure from extremely short localisation microscopy image sequences. Our more recent work has shown that the maximum achievable acquisition speed of single emitter localisation type techniques is fundamentally limited by the shape of the tagged portion of the sample itself. This can cause changes in acquisition time that vary by orders of magnitude, and do so non-uniformly across the sample structure. The errors this introduces are systematic, and hard to identify from the distribution of localisations alone. However a Random forest classifier, trained by the user on localisations which can be seen by eye to be clearly inaccurate, can quickly and accurately correct for this.