Benchmarking molecular feature attribution methods with activity cliffs

Feature attribution techniques are popular choices within the explainable artificial intelligence toolbox, as they can help elucidate which parts of the provided inputs used by an underlying supervised-learning method are considered relevant for a specific prediction. In the context of molecular design, these approaches typically involve the coloring of molecular graphs, whose presentation to medicinal chemists can be useful for making a decision of which compounds to synthesize or prioritize. The consistency of the highlighted moieties alongside expert background knowledge is expected to contribute to the understanding of machine-learning models in drug design. Quantitative evaluation of such coloring approaches, however, has so far been limited to substructure identification tasks. We here present an approach that is based on maximum common substructure algorithms applied to experimentally-determined activity cliffs. Using the proposed benchmark, we found that molecule coloring approaches in conjunction with classical machine-learning models tend to outperform more modern, deep-learning-based alternatives. However, none of the tested feature attribution methods sufficiently and consistently generalized when confronted with unseen examples.

Multi-taper method based substructure identification for shear structures

Quickly and accurately identifying structural status after natural disasters plays crucial roles in disaster rescue. Previously, the authors developed a substructure identification method for shear structures, which uses the frequency responses of short structural acceleration responses to estimate structural parameters inductively. However, the numerical studies found that the method could only provide moderately accurate results. In this paper, a thorough uncertainty analysis is performed to reveal the key factors that influence its identification accuracy. Based on these results, a new substructure method is proposed herein, which utilizes the cross power spectrum densities of structural responses, estimated by the multi-taper method, to formulate substructure identification problems. The error analysis is also conducted for the multi-taper method based method, explaining why this method can significantly improve identification accuracy, compared with the frequency response based method. Moreover, although the multi-taper method based method is originally derived based on stationary structural responses, a further analysis shows that it can be extended to non-stationary responses, greatly broadening the method’s application range. Finally, the simulation study of a 20-story shear structure and the shake table tests on a three-story bench-scaled structure are conducted, which verified that the proposed multi-taper method based method indeed significantly improves the substructure identification accuracy.

EBSD in Antarctic and Greenland Ice

EBSD provides information for the characterization of subgrain boundary types and dislocation activity during deformation. EBSD microstructure in combination with light microscopy measurements from ice core material from Antarctica (EPICA-DML deep ice core) and Greenland (NEEM deep ice core) are presented and interpreted regarding substructure identification and characterization. Electron backscattered diffraction (EBSD) analyses suggest that a large portion of edge dislocations with slip systems basal <a> gliding on the basal plane were involved in forming subgrain boundaries. However, an almost equal number of tilt subgrain boundaries developed, involving dislocations gliding on non basal planes (prism <c> or prism <c+a> slip). A few subgrain boundaries involving prism <a> edge dislocation glide, as well as boundaries involving basal <a> twist dislocations, were also identified. The finding that subgrain boundaries occur, made up of dislocations gliding on non-basal planes, are as frequent as basal plane slip systems, is surprising. These findings are expected to have an impact on the discussion of rate-controlling processes for the ice flow descriptions of large ice masses with respect to sea-level evolution. For subgrain boundaries not related to the crystallography of the host grain alternative formation processes are discussed.