Spur Gear: The spur gear has a helix angle of 0°. Worm Gear: Worm gears are found in right angle gearboxes. They “turn a corner.” Helical Gear: This is an angle toothed gear. Bevel Gear: Bevel gears tend to have a lower ratio and run at a higher efficiency than worm gears.
Classification of Gears
Gears are classified into three categories: parallel axis gears, intersecting axis gears, and nonparallel and nonintersecting axis gears. Spur gears and helical gears fall under parallel axis gears, while bevel gears fall under intersecting axis gears. Screw or crossed helical, worm gears, and hypoid gears are categorized as nonparallel and nonintersecting axis gears.
Gears use the principle of mechanical advantage, defined as the ratio of output force to input force in a system. In the case of gears, the mechanical advantage is determined by the gear ratio, which is the ratio of the final gear's speed to the initial gear's speed. For more details, kindly visit Kexin.
Predicting Transcriptional Outcomes of Novel Multigene Perturbations
Knowledge-informed Deep Learning of Perturbation Effects
GEARS is a deep learning-based model that predicts the gene expression outcome when a set of one or more genes is perturbated (perturbation set). Given unperturbed single-cell gene expression data along with the applied perturbation set, the model outputs the transcriptional state of the cell following the perturbation.
Fig. 1: GEARS combines prior knowledge with deep learning to predict postperturbation gene expression.
a, Problem formulation: given unperturbed gene expression (green) and applied perturbation (red), predict the gene expression outcome (purple). Each box corresponds to an individual gene. Arrows indicate change in expression. b, GEARS model architecture. (i) For each gene in the unperturbed state, GEARS initializes a gene embedding vector (green) and a gene perturbation embedding vector (red) (ii). These embedding vectors are assigned as node features in the gene relationship graph and the perturbation relationship graph (iii). A GNN is used to combine information between neighbors in each graph. Each resulting gene embedding is summed with the perturbation embedding of each perturbation in the perturbation set (iv). The output is combined across all genes using the cross-gene layer and fed into gene-specific output layers (v). The final result is postperturbation gene expression; MLP, multilayer perceptron.
GEARS introduces a novel method of representing each gene and its perturbation using distinct multidimensional embeddings. Each gene's embedding is optimized during training to encode key traits of that gene. This dual embedding approach enhances the model's capability to capture gene-specific heterogeneity in perturbation responses. Each gene’s embedding is sequentially combined with the perturbation embedding of each gene in the perturbation set to predict the postperturbation gene expression state, conditioned on a 'cross-gene' embedding vector containing transcriptome-wide information for each cell.
GEARS can predict outcomes for perturbation sets that include genes without any available experimental perturbation data. It incorporates prior knowledge of gene-gene relationships through a gene coexpression knowledge graph for gene embeddings and a Gene Ontology (GO)-derived knowledge graph for gene perturbation embeddings. This approach relies on two biological intuitions: (i) genes with similar expression patterns are likely to respond similarly to perturbations, and (ii) genes involved in similar pathways are likely to impact the same genes upon perturbation. Depending on the gene set of interest, different knowledge graphs, such as large context-specific networks, may be more suitable. GEARS utilizes a graph neural network (GNN) architecture to incorporate this graph-based inductive bias.
GEARS was evaluated on single-gene perturbations where data for certain genes were excluded from the training phase. This evaluation included data from two genetic perturbation screens: 1,543 perturbations in RPE-1 cells and 1,092 perturbations in K562 cells, each with over 170,000 measured cells. These screens utilized the Perturb-seq assay, a technique combining pooled screen with single-cell RNA sequencing of the entire transcriptome for each cell. GEARS was trained on each dataset separately. For performance evaluation, two baseline models were used: one assuming no change due to perturbation (no perturbation) and another inferring a gene regulatory network and propagating perturbation effects linearly (adapted from CellOracle).
Model performance was tested using mean squared error (m.s.e.) and Pearson correlation between predicted and true postperturbation gene expressions for the held-out dataset. GEARS significantly outperformed all baseline models with more than 30-50% improvement in m.s.e. and over two times better performance in Pearson correlation for both cell lines. Additionally, GEARS showed a marked improvement in capturing the correct direction of gene expression changes following perturbation, indicating a more accurate representation of regulatory relationships. The superior performance of GEARS was consistent across multiple datasets, including a genome-wide perturbation screen, and scaled more effectively to large datasets than traditional gene regulatory network-based methods. Beyond transcription levels, GEARS identified gene groups inducing similar transcriptional responses to perturbation, even without training data.
Comments
All Comments (0)