Assessing the utility of CASP14 models for molecular replacement

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Assessing protein models for molecular replacement (MR) is crucial for structure determination.
• Previous CASP assessments relied on metrics requiring diffraction data and specific crystal forms, limiting scope.
• The log-likelihood-gain (LLG) score in CASP10 enabled broader assessment but still required diffraction data.

Purpose of the Study:

• To introduce a novel metric, relative-expected-LLG (reLLG), for assessing protein models in molecular replacement.
• To develop a data-independent method for evaluating model utility across various structural biology sources (X-ray, NMR, cryo-EM).

Main Methods:

• Development of the relative-expected-LLG (reLLG) metric, independent of diffraction data and crystal form.
• Calibration of reLLG against the LLG metric using CASP14 targets.
• Evaluation of the impact of coordinate error estimates and model refinement on MR success.

Main Results:

• The reLLG metric is a robust measure for ranking protein models and assessment groups.
• Accurate coordinate error estimates significantly enhance the value of predicted models for MR.
• Refinement of initial models often transforms them into successful MR search models.
• AlphaFold2 models demonstrate superior performance for molecular replacement phasing compared to other methods.

Conclusions:

• The reLLG metric offers a versatile and data-independent approach for evaluating protein models for molecular replacement.
• Accurate error estimation and model refinement are key to improving the utility of predicted models.
• AlphaFold2 models represent a significant advancement, outperforming existing strategies for MR phasing.