Correcting gradient-based interpretations of deep neural networks for genomics.

Post hoc attribution methods can provide insights into the learned patterns from deep neural networks (DNNs) trained on high-throughput functional genomics data. However, in practice, their resultant attribution maps can be challenging to interpret due to spurious importance scores for seemingly arbitrary nucleotides. Here, we identify a previously overlooked attribution noise source that arises from how DNNs handle one-hot encoded DNA. We demonstrate this noise is pervasive across various genomic DNNs and introduce a statistical correction that effectively reduces it, leading to more reliable attribution maps. Our approach represents a promising step towards gaining meaningful insights from DNNs in regulatory genomics.

ResidualBind: Uncovering Sequence-Structure Preferences of RNA-Binding Proteins with Deep Neural Networks.

Deep neural networks have demonstrated improved performance at predicting sequence specificities of DNA- and RNA-binding proteins. However, it remains unclear why they perform better than previous methods that rely on k-mers and position weight matrices. Here, we highlight a recent deep learning-based software package, called ResidualBind, that analyzes RNA-protein interactions using only RNA sequence as an input feature and performs global importance analysis for model interpretability. We discuss practical considerations for model interpretability to uncover learned sequence motifs and their secondary structure preferences.

Selecting deep neural networks that yield consistent attribution-based interpretations for genomics.

Deep neural networks (DNNs) have advanced our ability to take DNA primary sequence as input and predict a myriad of molecular activities measured via high-throughput functional genomic assays. Post hoc attribution analysis has been employed to provide insights into the importance of features learned by DNNs, often revealing patterns such as sequence motifs. However, attribution maps typically harbor spurious importance scores to an extent that varies from model to model, even for DNNs whose predictions generalize well. Thus, the standard approach for model selection, which relies on performance of a held-out validation set, does not guarantee that a high-performing DNN will provide reliable explanations. Here we introduce two approaches that quantify the consistency of important features across a population of attribution maps; consistency reflects a qualitative property of human interpretable attribution maps. We employ the consistency metrics as part of a multivariate model selection framework to identify models that yield high generalization performance and interpretable attribution analysis. We demonstrate the efficacy of this approach across various DNNs quantitatively with synthetic data and qualitatively with chromatin accessibility data.

Global importance analysis: An interpretability method to quantify importance of genomic features in deep neural networks.

Deep neural networks have demonstrated improved performance at predicting the sequence specificities of DNA- and RNA-binding proteins compared to previous methods that rely on k-mers and position weight matrices. To gain insights into why a DNN makes a given prediction, model interpretability methods, such as attribution methods, can be employed to identify motif-like representations along a given sequence. Because explanations are given on an individual sequence basis and can vary substantially across sequences, deducing generalizable trends across the dataset and quantifying their effect size remains a challenge. Here we introduce global importance analysis (GIA), a model interpretability method that quantifies the population-level effect size that putative patterns have on model predictions. GIA provides an avenue to quantitatively test hypotheses of putative patterns and their interactions with other patterns, as well as map out specific functions the network has learned. As a case study, we demonstrate the utility of GIA on the computational task of predicting RNA-protein interactions from sequence. We first introduce a convolutional network, we call ResidualBind, and benchmark its performance against previous methods on RNAcompete data. Using GIA, we then demonstrate that in addition to sequence motifs, ResidualBind learns a model that considers the number of motifs, their spacing, and sequence context, such as RNA secondary structure and GC-bias.

Multiple tipping points and optimal repairing in interacting networks.

Systems composed of many interacting dynamical networks-such as the human body with its biological networks or the global economic network consisting of regional clusters-often exhibit complicated collective dynamics. Three fundamental processes that are typically present are failure, damage spread and recovery. Here we develop a model for such systems and find a very rich phase diagram that becomes increasingly more complex as the number of interacting networks increases. In the simplest example of two interacting networks we find two critical points, four triple points, ten allowed transitions and two 'forbidden' transitions, as well as complex hysteresis loops. Remarkably, we find that triple points play the dominant role in constructing the optimal repairing strategy in damaged interacting systems. To test our model, we analyse an example of real interacting financial networks and find evidence of rapid dynamical transitions between well-defined states, in agreement with the predictions of our model.

Predicting the Lifetime of Dynamic Networks Experiencing Persistent Random Attacks.

Estimating the critical points at which complex systems abruptly flip from one state to another is one of the remaining challenges in network science. Due to lack of knowledge about the underlying stochastic processes controlling critical transitions, it is widely considered difficult to determine the location of critical points for real-world networks, and it is even more difficult to predict the time at which these potentially catastrophic failures occur. We analyse a class of decaying dynamic networks experiencing persistent failures in which the magnitude of the overall failure is quantified by the probability that a potentially permanent internal failure will occur. When the fraction of active neighbours is reduced to a critical threshold, cascading failures can trigger a total network failure. For this class of network we find that the time to network failure, which is equivalent to network lifetime, is inversely dependent upon the magnitude of the failure and logarithmically dependent on the threshold. We analyse how permanent failures affect network robustness using network lifetime as a measure. These findings provide new methodological insight into system dynamics and, in particular, of the dynamic processes of networks. We illustrate the network model by selected examples from biology, and social science.

Spontaneous curvature as a regulator of the size of virus capsids.

We investigate the physical reasons underlying the high monodispersity of empty virus capsids assembled in thermodynamical equilibrium in conditions of favorable pH and ionic strength. We propose that the high fidelity of the assembly results from the effective spontaneous curvature of the viral protein assemblies and the corresponding bending rigidity that penalizes curvatures which are larger and smaller from the spontaneous one. On the example of hepatitis B virus, which has been thoroughly studied experimentally in the context of interest to us, we estimate the magnitude of bending rigidity that is needed to suppress the appearance of aberrant capsid structures (approximately 60k(B)T). Our approach also demonstrates that the aberrant capsids that can be classified within the Caspar-Klug framework are in most circumstances likely to be smaller from the regular ones, in agreement with the experimental findings.