Neuron-level explainable AI for Alzheimer's Disease assessment from fundus images.

Alzheimer's Disease (AD) is a progressive neurodegenerative disease and the leading cause of dementia. Early diagnosis is critical for patients to benefit from potential intervention and treatment. The retina has emerged as a plausible diagnostic site for AD detection owing to its anatomical connection with the brain. However, existing AI models for this purpose have yet to provide a rational explanation behind their decisions and have not been able to infer the stage of the disease's progression. Along this direction, we propose a novel model-agnostic explainable-AI framework, called Granu la ̲ r Neuron-le v ̲ el Expl a ̲ iner (LAVA), an interpretation prototype that probes into intermediate layers of the Convolutional Neural Network (CNN) models to directly assess the continuum of AD from the retinal imaging without the need for longitudinal or clinical evaluations. This innovative approach aims to validate retinal vasculature as a biomarker and diagnostic modality for evaluating Alzheimer's Disease. Leveraged UK Biobank cognitive tests and vascular morphological features demonstrate significant promise and effectiveness of LAVA in identifying AD stages across the progression continuum.

OnDev-LCT: On-Device Lightweight Convolutional Transformers towards federated learning.

Federated learning (FL) has emerged as a promising approach to collaboratively train machine learning models across multiple edge devices while preserving privacy. The success of FL hinges on the efficiency of participating models and their ability to handle the unique challenges of distributed learning. While several variants of Vision Transformer (ViT) have shown great potential as alternatives to modern convolutional neural networks (CNNs) for centralized training, the unprecedented size and higher computational demands hinder their deployment on resource-constrained edge devices, challenging their widespread application in FL. Since client devices in FL typically have limited computing resources and communication bandwidth, models intended for such devices must strike a balance between model size, computational efficiency, and the ability to adapt to the diverse and non-IID data distributions encountered in FL. To address these challenges, we propose OnDev-LCT: Lightweight Convolutional Transformers for On-Device vision tasks with limited training data and resources. Our models incorporate image-specific inductive biases through the LCT tokenizer by leveraging efficient depthwise separable convolutions in residual linear bottleneck blocks to extract local features, while the multi-head self-attention (MHSA) mechanism in the LCT encoder implicitly facilitates capturing global representations of images. Extensive experiments on benchmark image datasets indicate that our models outperform existing lightweight vision models while having fewer parameters and lower computational demands, making them suitable for FL scenarios with data heterogeneity and communication bottlenecks.

QuTIE: quantum optimization for target identification by enzymes.

Summary: Target identification by enzymes (TIE) problem aims to identify the set of enzymes in a given metabolic network, such that their inhibition eliminates a given set of target compounds associated with a disease while incurring minimum damage to the rest of the compounds. This is a NP-hard problem, and thus optimal solutions using classical computers fail to scale to large metabolic networks. In this article, we develop the first quantum optimization solution, called QuTIE (quantum optimization for target identification by enzymes), to this NP-hard problem. We do that by developing an equivalent formulation of the TIE problem in quadratic unconstrained binary optimization form. We then map it to a logical graph, and embed the logical graph on a quantum hardware graph. Our experimental results on 27 metabolic networks from Escherichia coli, Homo sapiens, and Mus musculus show that QuTIE yields solutions that are optimal or almost optimal. Our experiments also demonstrate that QuTIE can successfully identify enzyme targets already verified in wet-lab experiments for 14 major disease classes. Availability and implementation: Code and sample data are available at: https://github.com/ngominhhoang/Quantum-Target-Identification-by-Enzymes.

Auditing the Sensitivity of Graph-based Ranking with Visual Analytics.

Graph mining plays a pivotal role across a number of disciplines, and a variety of algorithms have been developed to answer who/what type questions. For example, what items shall we recommend to a given user on an e-commerce platform? The answers to such questions are typically returned in the form of a ranked list, and graph-based ranking methods are widely used in industrial information retrieval settings. However, these ranking algorithms have a variety of sensitivities, and even small changes in rank can lead to vast reductions in product sales and page hits. As such, there is a need for tools and methods that can help model developers and analysts explore the sensitivities of graph ranking algorithms with respect to perturbations within the graph structure. In this paper, we present a visual analytics framework for explaining and exploring the sensitivity of any graph-based ranking algorithm by performing perturbation-based what-if analysis. We demonstrate our framework through three case studies inspecting the sensitivity of two classic graph-based ranking algorithms (PageRank and HITS) as applied to rankings in political news media and social networks.

Construction of Signaling Pathways with RNAi Data and Multiple Reference Networks.

Signaling networks are involved in almost all major diseases such as cancer. As a result of this, understanding how signaling networks function is vital for finding new treatments for many diseases. Using gene knockdown assays such as RNA interference (RNAi) technology, many genes involved in these networks can be identified. However, determining the interactions between these genes in the signaling networks using only experimental techniques is very challenging, as performing extensive experiments is very expensive and sometimes, even impractical. Construction of signaling networks from RNAi data using computational techniques have been proposed as an alternative way to solve this challenging problem. However, the earlier approaches are either not scalable to large scale networks, or their accuracy levels are not satisfactory. In this study, we integrate RNAi data given on a target network with multiple reference signaling networks and phylogenetic trees to construct the topology of the target signaling network. In our work, the network construction is considered as finding the minimum number of edit operations on given multiple reference networks, in which their contributions are weighted by their phylogenetic distances to the target network. The edit operations on the reference networks lead to a target network that satisfies the RNAi knockdown observations. Here, we propose two new reference-based signaling network construction methods that provide optimal results and scale well to large-scale signaling networks of hundreds of components. We compare the performance of these approaches to the state-of-the-art reference-based network construction method SiNeC on synthetic, semi-synthetic, and real datasets. Our analyses show that the proposed methods outperform SiNeC method in terms of accuracy. Furthermore, we show that our methods function well even if evolutionarily distant reference networks are used. Application of our methods to the Apoptosis and Wnt signaling pathways recovers the known protein-protein interactions and suggests additional relevant interactions that can be tested experimentally.

Dynamic social community detection and its applications.

Community structure is one of the most commonly observed features of Online Social Networks (OSNs) in reality. The knowledge of this feature is of great advantage: it not only provides helpful insights into developing more efficient social-aware solutions but also promises a wide range of applications enabled by social and mobile networking, such as routing strategies in Mobile Ad Hoc Networks (MANETs) and worm containment in OSNs. Unfortunately, understanding this structure is very challenging, especially in dynamic social networks where social interactions are evolving rapidly. Our work focuses on the following questions: How can we efficiently identify communities in dynamic social networks? How can we adaptively update the network community structure based on its history instead of recomputing from scratch? To this end, we present Quick Community Adaptation (QCA), an adaptive modularity-based framework for not only discovering but also tracing the evolution of network communities in dynamic OSNs. QCA is very fast and efficient in the sense that it adaptively updates and discovers the new community structure based on its history together with the network changes only. This flexible approach makes QCA an ideal framework applicable for analyzing large-scale dynamic social networks due to its lightweight computing-resource requirement. To illustrate the effectiveness of our framework, we extensively test QCA on both synthesized and real-world social networks including Enron, arXiv e-print citation, and Facebook networks. Finally, we demonstrate the applicability of QCA in real applications: (1) A social-aware message forwarding strategy in MANETs, and (2) worm propagation containment in OSNs. Competitive results in comparison with other methods reveal that social-based techniques employing QCA as a community detection core outperform current available methods.

Efficient non-unique probes selection algorithms for DNA microarray.

BACKGROUND: Temperature and salt concentration are very helpful experimental conditions for a probe to hybridize uniquely to its intended target. In large families of closely related target sequences, the high degree of similarity makes it impossible to find a unique probe for every target. We studied how to select a minimum set of non-unique probes to identify the presence of at most d targets in a sample where each non-unique probe can hybridize to a set of targets. RESULTS: We proposed efficient algorithms based on Integer Linear Programming to select a minimum number of non-unique probes using d-disjunct matrices. Our non-unique probes selection can also identify up to d targets in a sample with at most k experimental errors. The decoding complexity of our algorithms is as simple as O(n). The experimental results show that the decoding time is much faster than that of the methods using d-separable matrices while running time and solution size are comparable. CONCLUSIONS: Since finding unique probes is often not easy, we make use of non-unique probes. Minimizing the number of non-unique probes will result in a smaller DNA microarry design which leads to a smaller chip and considerable reduction of cost. While minimizing the probe set, the decoding ability should not be diminished. Our non-unique probes selection algorithms can identify up to d targets with error tolerance and the decoding complexity is O(n).

Decoding algorithms in pooling designs with inhibitors and error-tolerance.

Pooling designs are used in DNA library screening to efficiently distinguish positive from negative clones, which is fundamental for studying gene functions and many other biological applications. One challenge is to design decoding algorithms for determining whether a clone is positive based on the test outcomes and a binary matrix representing the pools. This is more difficult in practice due to errors in biological experiments. More challenging still is a third category of clones called 'inhibitors' whose effect is to neutralise positives. We present a novel decoding algorithm identifying all positive clones in the presence of inhibitors and experimental errors.

Protein-protein interaction and group testing in bipartite graphs.

The interactions between bait proteins and prey proteins are often critical in many biological processes, such as the formation of macromolecular complexes and the transduction of signals in biological pathways. Thus, identifying all protein-protein interactions is an important task in those processes, which can be formulated as a group testing problem in bipartite graphs. In this paper, we take the advantages of the characteristics of bipartite graphs and present two nonadaptive algorithms for this problem. Furthermore, we illustrate a generalisation of our solution in a more general case.