A total of 2,000 HVGs were selected on the basis of the reference datasets. We quantified the quality of the data integration using the following metrics from the scIB (v 1.0.0) package and Luecken et al. Cell type ASW (average silhouette width), isolated label F1, isolated label silhouette, NMI (normalized mutual information) and ARI (adjusted Rand index) were used as biological conservation metrics. To quantify batch mixing we used PC regression, graph connectivity and batch ASW.
The structure of plain woven C/C composite laminate exhibits heterogeneous multi-scale characteristics. The macroscale structure is a C/C composite laminate for finger seals, and the mesoscale structure is a RWC (representative woven cell) composed of orthogonal plain woven carbon fiber yarns and carbon matrix. The micro-scale structure is the RVE (representative volume element) multi-scale analysis inside the carbon fiber yarn, which is composed of a large number of randomly distributed carbon fibers and carbon matrix. The growth of multiscale modeling in the industrial sector was primarily due to financial motivations. From the DOE national labs perspective, the shift from large-scale systems experiments mentality occurred because of the 1996 Nuclear Ban Treaty.
The overall integration score is a weighted average of the average batch mixing score and the average biological conservation score, with weight 0.4 and 0.6 respectively. To quantify label transfer accuracy we used the weighted averaged and macro-averaged F1 score. Radial dynamic stiffness of finger beams with various displacement amplitudes. In this paper, a piece of plain woven C/C composite of 100 × 100 × 2.1 mm was selected and the corresponding finger seal specimen was processed by a laser cutting machine tool.
We found that scPoli outperformed the next best-performing model (scANVI) by 5.06% in data integration (Fig. 2a). When we looked separately at batch correction and biological conservation metrics, we observed that scPoli preserved biologically meaningful signals better than other methods. To understand whether the improvements stemmed from the use of condition embeddings or from the inclusion of the prototype loss, we benchmarked two variants of our model.