Implementing Novel Data Analysis Methods to Enhance Biophysical Studies

Abstract

Observing and quantifying the interactions of proteins as they perform their biological functions in living cells is of high importance in modern biophysical research. Serial femtosecond crystallography (SFX) and time-resolved serial femtosecond crystallography (TR-SFX) have been successful in obtaining protein structures at near-atomic spatial resolutions and ultrafast temporal resolutions. However, this method of obtaining protein structures presents data analysis challenges. In typical SFX experiments, the microcrystals are streamed through the X-ray exposure path, and each X-ray pulse can only interact with a crystal for a short duration of time before it is destroyed. Hence, each collected diffraction pattern is only a small portion of the full diffraction information needed to fully characterize the structure of the proteins in the crystal. Therefore, the collected data is highly incomplete. Further, the data may also suffer from issues like noise and timing uncertainty. Classically, SFX data has been analyzed by merging all the collected data to obtain a single structure from the whole dataset. Time-resolved datasets have, on the other hand, been analyzed by first grouping data into temporal bins before averaging the data in each bin to obtain a time series of distinct structures. While this method has yielded structural information for many proteins, it loses finer resolution information. In this work, we employ a machine learning algorithm known as nonlinear Laplacian spectral analysis, or NLSA, to fill the data analysis gaps left by simpler averaging methods. To test the effectiveness of NLSA, we simulated sets of diffraction data for photoactive yellow protein (PYP) suffering from noise, incompleteness and timing uncertainty. With this, we demonstrate that NLSA is an effective algorithm for overcoming noise and timing uncertainty and can recover useful structural and dynamical information from TR-SFX experiments. Other ways of studying proteins have also proven very fruitful, specifically when studying phenomena like protein-protein interactions. One such method is the use of fluorescence microscopy paired with Förster resonance energy transfer, or FRET. FRET is the non-radiative transfer of energy from an excited donor fluorescent molecule to a nearby acceptor. Since FRET is highly sensitive to the separation distance between donor and acceptor molecules, it is a natural choice in quantifying protein-protein interactions in cells. In biological studies, fluorescent proteins are commonly used, which serve as donors and acceptors, and can be tagged to other proteins of interest to quantify interaction. The practice of FRET spectrometry has revealed geometrical properties such as the quaternary structure of proteins but has been limited to using spectrally resolved instruments. In this work, we implement an alternative method of processing data from time-resolved fluorescence decay signal, as would be obtained in fluorescence lifetime imaging microscopy (FLIM) studies. This new approach, dubbed tiFRET, involves integration of the fluorescence decay signal instead of fitting with exponential functions. This methodology may allow users of FLIM to perform FRET spectrometry, which may broaden the capabilities of FLIM practitioners.