Dictionary learning allows model-free pseudotime estimation of transcriptomic data

Background Pseudotime estimation from dynamic single-cell transcriptomic data enables characterisation and understanding of the underlying processes, for example developmental processes. Various pseudotime estimation methods have been proposed during the last years. Typically, these methods start with a dimension reduction step because the low-dimensional representation is usually easier to analyse. Approaches such as PCA, ICA or t-SNE belong to the most widely used methods for dimension reduction in pseudotime estimation methods. However, these methods usually make assumptions on the derived dimensions, which can result in important dataset properties being missed. In this paper, we suggest a new dictionary learning based approach, dynDLT, for dimension reduction and pseudotime estimation of dynamic transcriptomic data. Dictionary learning is a matrix factorisation approach that does not restrict the dependence of the derived dimensions. To evaluate the performance, we conduct a large simulation study and analyse 8 real-world datasets. Results The simulation studies reveal that firstly, dynDLT preserves the simulated patterns in low-dimension and the pseudotimes can be derived from the low-dimensional representation. Secondly, the results show that dynDLT is suitable for the detection of genes exhibiting the simulated dynamic patterns, thereby facilitating the interpretation of the compressed representation and thus the dynamic processes. For the real-world data analysis, we select datasets with samples that are taken at different time points throughout an experiment. The pseudotimes found by dynDLT have high correlations with the experimental times. We compare the results to other approaches used in pseudotime estimation, or those that are method-wise closely connected to dictionary learning: ICA, NMF, PCA, t-SNE, and UMAP. DynDLT has the best overall performance for the simulated and real-world datasets. Conclusions We introduce dynDLT, a method that is suitable for pseudotime estimation. Its main advantages are: (1) It presents a model-free approach, meaning that it does not restrict the dependence of the derived dimensions; (2) Genes that are relevant in the detected dynamic processes can be identified from the dictionary matrix; (3) By a restriction of the dictionary entries to positive values, the dictionary atoms are highly interpretable.

Random Compressed Coding with Neurons

Classical models of efficient coding in neurons assume simple mean responses--'tuning curves'--such as bell shaped or monotonic functions of a stimulus feature. Real neurons, however, can be more complex: grid cells, for example, exhibit periodic responses which impart the neural population code with high accuracy. But do highly accurate codes require fine tuning of the response properties? We address this question with the use of a benchmark model: a neural network with random synaptic weights which result in output cells with irregular tuning curves. Irregularity enhances the local resolution of the code but gives rise to catastrophic, global errors. For optimal smoothness of the tuning curves, when local and global errors balance out, the neural network compresses information from a high-dimensional representation to a low-dimensional one, and the resulting distributed code achieves exponential accuracy. An analysis of recordings from monkey motor cortex points to such 'compressed efficient coding'. Efficient codes do not require a finely tuned design--they emerge robustly from irregularity or randomness.

Projection Path Explorer: Exploring Visual Patterns in Projected Decision-making Paths

In problem-solving, a path towards a solutions can be viewed as a sequence of decisions. The decisions, made by humans or computers, describe a trajectory through a high-dimensional representation space of the problem. By means of dimensionality reduction, these trajectories can be visualized in lower-dimensional space. Such embedded trajectories have previously been applied to a wide variety of data, but analysis has focused almost exclusively on the self-similarity of single trajectories. In contrast, we describe patterns emerging from drawing many trajectories—for different initial conditions, end states, and solution strategies—in the same embedding space. We argue that general statements about the problem-solving tasks and solving strategies can be made by interpreting these patterns. We explore and characterize such patterns in trajectories resulting from human and machine-made decisions in a variety of application domains: logic puzzles (Rubik’s cube), strategy games (chess), and optimization problems (neural network training). We also discuss the importance of suitably chosen representation spaces and similarity metrics for the embedding.

Predicting Pseudogene–miRNA Associations Based on Feature Fusion and Graph Auto-Encoder

Pseudogenes were originally regarded as non-functional components scattered in the genome during evolution. Recent studies have shown that pseudogenes can be transcribed into long non-coding RNA and play a key role at multiple functional levels in different physiological and pathological processes. microRNAs (miRNAs) are a type of non-coding RNA, which plays important regulatory roles in cells. Numerous studies have shown that pseudogenes and miRNAs have interactions and form a ceRNA network with mRNA to regulate biological processes and involve diseases. Exploring the associations of pseudogenes and miRNAs will facilitate the clinical diagnosis of some diseases. Here, we propose a prediction model PMGAE (Pseudogene–MiRNA association prediction based on the Graph Auto-Encoder), which incorporates feature fusion, graph auto-encoder (GAE), and eXtreme Gradient Boosting (XGBoost). First, we calculated three types of similarities including Jaccard similarity, cosine similarity, and Pearson similarity between nodes based on the biological characteristics of pseudogenes and miRNAs. Subsequently, we fused the above similarities to construct a similarity profile as the initial representation features for nodes. Then, we aggregated the similarity profiles and associations of nodes to obtain the low-dimensional representation vector of nodes through a GAE. In the last step, we fed these representation vectors into an XGBoost classifier to predict new pseudogene–miRNA associations (PMAs). The results of five-fold cross validation show that PMGAE achieves a mean AUC of 0.8634 and mean AUPR of 0.8966. Case studies further substantiated the reliability of PMGAE for mining PMAs and the study of endogenous RNA networks in relation to diseases.

Is it left or is it right? A classification approach for investigating hemispheric differences in low and high dimensionality

Hemispheric asymmetries, i.e., differences between the two halves of the brain, have extensively been studied with respect to both structure and function. Commonly employed pairwise comparisons between left and right are suitable for finding differences between the hemispheres, but they come with several caveats when assessing multiple asymmetries. What is more, they are not designed for identifying the characterizing features of each hemisphere. Here, we present a novel data-driven framework—based on machine learning-based classification—for identifying the characterizing features that underlie hemispheric differences. Using voxel-based morphometry data from two different samples (n = 226, n = 216), we separated the hemispheres along the midline and used two different pipelines: First, for investigating global differences, we embedded the hemispheres into a two-dimensional space and applied a classifier to assess if the hemispheres are distinguishable in their low-dimensional representation. Second, to investigate which voxels show systematic hemispheric differences, we employed two classification approaches promoting feature selection in high dimensions. The two hemispheres were accurately classifiable in both their low-dimensional (accuracies: dataset 1 = 0.838; dataset 2 = 0.850) and high-dimensional (accuracies: dataset 1 = 0.966; dataset 2 = 0.959) representations. In low dimensions, classification of the right hemisphere showed higher precision (dataset 1 = 0.862; dataset 2 = 0.894) compared to the left hemisphere (dataset 1 = 0.818; dataset 2 = 0.816). A feature selection algorithm in the high-dimensional analysis identified voxels that most contribute to accurate classification. In addition, the map of contributing voxels showed a better overlap with moderate to highly lateralized voxels, whereas conventional t test with threshold-free cluster enhancement best resembled the LQ map at lower thresholds. Both the low- and high-dimensional classifiers were capable of identifying the hemispheres in subsamples of the datasets, such as males, females, right-handed, or non-right-handed participants. Our study indicates that hemisphere classification is capable of identifying the hemisphere in their low- and high-dimensional representation as well as delineating brain asymmetries. The concept of hemisphere classifiability thus allows a change in perspective, from asking what differs between the hemispheres towards focusing on the features needed to identify the left and right hemispheres. Taking this perspective on hemispheric differences may contribute to our understanding of what makes each hemisphere special.

Bringing the German Pavilion Back Home

<p>This research investigates a correspondence between the architectural representational tool of drawing, and the translations of these into something recognised as ‘built’. It is fundamentally concerned around representation in architecture driven by the principles that our entire engagement with architecture is via representation. Architects do not produce buildings but produce images of buildings, and the role of two-dimensional representation plays a principal part in architecture. Architecture is always representational, and the more we engage with representation the more we might push the envelope with what we understand architecture to be. This thesis aims to establish within the contemporary discipline, what we understand about the responsibility of linear perspective as a representational tool. By understanding what lies behind the canon of perspective in architecture, this thesis questions whether the representation of conventional architecture could benefit from a new way of drawing linear perspective? The discovery of perspective during the Renaissance has influenced not only our way of representing architecture but also how we view, and therefore design it. It has become integrated with our understanding of architecture at an unconscious level. Architects no longer need control of projective geometry, and due to this cannot be critical of the system of representation or control its limits. This leads to mediate a shift in perspective, with the intention to generate a representation of new form. The motivation for this thesis was that from linear perspective, as it has done so for centuries, we can produce evocative and meaningful vocabularies that attempt to enrich architecture.</p>

Bringing the German Pavilion Back Home

<p>This research investigates a correspondence between the architectural representational tool of drawing, and the translations of these into something recognised as ‘built’. It is fundamentally concerned around representation in architecture driven by the principles that our entire engagement with architecture is via representation. Architects do not produce buildings but produce images of buildings, and the role of two-dimensional representation plays a principal part in architecture. Architecture is always representational, and the more we engage with representation the more we might push the envelope with what we understand architecture to be. This thesis aims to establish within the contemporary discipline, what we understand about the responsibility of linear perspective as a representational tool. By understanding what lies behind the canon of perspective in architecture, this thesis questions whether the representation of conventional architecture could benefit from a new way of drawing linear perspective? The discovery of perspective during the Renaissance has influenced not only our way of representing architecture but also how we view, and therefore design it. It has become integrated with our understanding of architecture at an unconscious level. Architects no longer need control of projective geometry, and due to this cannot be critical of the system of representation or control its limits. This leads to mediate a shift in perspective, with the intention to generate a representation of new form. The motivation for this thesis was that from linear perspective, as it has done so for centuries, we can produce evocative and meaningful vocabularies that attempt to enrich architecture.</p>

PIPP: Improving peptide identity propagation using neural networks

Peptide identity propagation (PIP) can substantially reduce missing values in label-free mass spectrometry quantification by transferring peptides identified by tandem mass (MS/MS) spectra in one run to experimentally related runs where the peptides are not identified by MS/MS. The existing frameworks for matching identifications between runs perform peak tracing and propagation based on similarity of precursor features using only a limited number of dimensions available in MS1 data. These approaches do not produce accompanying confidence estimates and hence cannot filter probable false positive identity transfers. We introduce an embedding based PIP that uses a higher dimensional representation of MS1 measurements that is optimized to capture peptide identities using deep neural networks. We developed a propagation framework that works entirely on MaxQuant results. Current PIP workflows typically perform propagation mainly using two feature dimensions, and rely on deterministic tolerances for identification transfer. Our framework overcomes both these limitations while additionally assigning probabilities to each transferred identity. The proposed embedding approach enables quantification of the empirical false discovery rate (FDR) for peptide identification, while also increasing depth of coverage through coembedding the runs from the experiment with experimental libraries. In published datasets with technical and biological variability, we demonstrate that our method reduces missing values in MaxQuant results, maintains high quantification precision and accuracy, and low false transfer rate.