Reliability of structural models is important for mechanistic studies of cellular processes and for rational design of drugs and treatment. Cryo-EM single particle reconstruction (Cryo-EM SPR) is an expanding technique that can generate atomic models for structural biology and has the advantage that it does not require samples to be crystallized. Instead, atomic models are built based on maps obtained by averaging hundreds of thousands of weak images, with each image containing snapshots of individual macromolecules suspended in a thin layer of ice. Due to the intrinsic difficulties with averaging noisy images, cryo-EM maps frequently have limited resolution and may represent averages of multiple structural states. Building and rebuilding structural models reliably in such maps remains a challenge and the process is additionally complicated by the lack of established criteria for validating the quality of models built at low resolution. When groups of a few atoms are individually recognizable in maps, established methods easily create reliable models. However, it is problematic when only larger groups of atoms are individually recognizable, and this is referred to as low resolution. It has been observed that for resolutions as low as 5 , reliable models can be built (with effort) if the maps are highly precise. This observation is an important factor for our plan to develop and implement methods for automatic, comprehensive and accurate building of atomic models for such low resolution. To achieve this goal, in Aim 1 we will improve the low resolution quality of cryo-EM maps by computing maps so that they are corrected for physical effects currently not properly considered in map creation.
In Aim 2, we will develop and implement an integrative procedure for automatic model building at low resolution. It will combine a fast 6D search using a library of continuous and discontinuous fragments obtained through the data mining of known structures. The hypotheses obtained from the 6D search will be analyze for forming self- consistent groups in a multi-stage process. This will be followed by GPU-accelerated molecular dynamics computations restrained by experimental data. The output of dynamics will undergo an additional layer of data mining to identify and trigger corrections of problematic starting assumptions and to create a concise description of information present in a multitude of calculated structural states. The final step will involve an assessment of possible unresolved ambiguities by the experimenter, who may have additional relevant knowledge guiding the selection of a specific structural hypothesis. Our SBIR phase I proposal will result in two software modules that will be incorporated into a commercial solution for data processing, analysis, and validation in cryo-EM.
Cryo-EM single particle reconstruction (Cryo-EM SPR) can generate accurate models of macromolecules that help us understand cellular processes in health and disease at the molecular level. However, building accurate structural models based on cryo-EM data is still challenging due to insufficient detail in the maps frequently obtained from cryo-EM SPR experiments. We will develop methods to more efficiently use the low resolution parts of cryo-EM maps and combine these methods with new approaches for building high quality structural models, so that cryo-EM can enlarge the project space for which it is a prime technique.
