Proteins perform much of the work in living organisms, and consequently the development of efficient computational methods for protein representation is essential for advancing large-scale biological research. Most current approaches struggle to efficiently integrate the wealth of information contained in the protein sequence and structure. In this paper, we propose a novel framework for embedding protein graphs in geometric vector spaces, by learning an encoder function that preserves the structural distance between protein graphs. Utilizing Graph Neural Networks (GNNs) and Large Language Models (LLMs), the proposed framework generates structure- and sequence- aware protein representations. We demonstrate that our embeddings are successful in the task of comparing protein structures, while providing a significant speed-up compared to traditional approaches based on structural alignment. Our framework achieves remarkable results in the task of protein structure classification; in particular, when compared to other work, the proposed method shows an average F1-Score improvement of 26% on out-of-distribution (OOD) samples and of 32% when tested on samples coming from the same distribution as the training data. Our approach finds applications in areas such as drug prioritization, drug re-purposing, disease sub-type analysis and elsewhere.

Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) involves acquiring a sequence of MRIs during the administration of a contrast agent. Radiologists then aim to discern the contrast uptake differences between malignant and benign lesions for tumor classification. Regrettably, existing literature underutilizes the temporal structure inherent to DCEMRI time series, leading to tumor classifications based on individual instants rather than entire sequences. This research introduces two Graph Neural Network (GNN)-based methods designed to aggregate information from multiple instants within the DCE-MRI sequence. Each lesion undergoes manual segmentation, and radiomic features are individually extracted from each time instant of the DCE-MRI sequence. Two graph construction methodologies are proposed: (i) a fully connected graph topology, aiming to represent each temporal instant as a node in a graph; (ii) a multiplex network, named MUGI-MRI (MUltiplex Graph neural network for Integration of MRI), where each layer identifies an instant of the DCE-MRI sequence. MUGI-MRI achieves an AUROC of 0.8017 ± 0.1146, showcasing promising performance in lesion classification. In addition to improving upon current state-of-the-art, the integration capability of MUGIMRI addresses the problem of imbalance between sensitivity and specificity, which affects numerous studies in the realm of DCE-MRI. Our findings strongly indicate that the aggregation of information across all time instants is pivotal for enhancing the diagnostic process, and vastly superior to a simplistic instant-wise analysis. While applied to MRI sequences, our approach can be extended to general problems of multimodal data integration.

Machine learning systems based on minimizing average error have been shown to perform inconsistently across notable subsets of the data, which is not exposed by a low average error for the entire dataset. In consequential social and economic applications, where data represent people, this can lead to discrimination of underrepresented gender and ethnic groups. Distributionally Robust Optimization (DRO) seemingly addresses this problem by minimizing the worst expected risk across subpopulations. We establish theoretical results that clarify the relation between DRO and the optimization of the same loss averaged on an adequately weighted training dataset. A practical implication of our results is that neither DRO nor curating the training set should be construed as a complete solution for bias mitigation.

Mahjong is a multi-player imperfect-information game with challenging features for AI research. Sanma, being a 3-player variant of Japanese Riichi Mahjong, possesses unique characteristics and a more aggressive playing style than the 4-player game. It is thus challenging and of research interest in its own right, but has not been explored. We present Meowjong, the first ever AI for Sanma using deep reinforcement learning (RL). We define a 2-dimensional data structure for encoding the observable information in a game. We pre-train 5 convolutional neural networks (CNNs) for Sanma’s 5 actions—discard, Pon, Kan, Kita and Riichi, and enhance the major (discard) action’s model via self-play reinforcement learning. Meowjong demonstrates potential for becoming the state-of-the-art in Sanma, by achieving test accuracies comparable with AIs for 4-player Mahjong through supervised learning, and gaining a significant further enhancement from reinforcement learning.

Modern theorem provers utilise a wide array of heuristics to control the search space explosion, thereby requiring optimisation of a large set of parameters. An exhaustive search in this multi-dimensional parameter space is intractable in most cases, yet the performance of the provers is highly dependent on the parameter assignment. In this work, we introduce a principled probabilistic framework for heuristic optimisation in theorem provers. We present results using a heuristic for premise selection and the Archive of Formal Proofs (AFP) as a case study.

Motivation: Protein-protein interactions (PPI) play a crucial role in our understanding of protein function and biological processes. The standardization and recording of experimental findings is increasingly stored in ontologies, with the Gene Ontology (GO) being one of the most successful projects. Several PPI evaluation algorithms have been based on the application of probabilistic frameworks or machine learning algorithms to GO properties. Here, we introduce a new training set design and machine learning based approach that combines dependent heterogeneous protein annotations from the entire ontology to evaluate putative co-complex protein interactions determined by empirical studies. Results: PPI annotations are built combinatorically using corresponding GO terms and InterPro annotation. We use a S.cerevisiae high-confidence complex dataset as a positive training set. A series of classifiers based on Maximum Entropy and support vector machines (SVMs), each with a composite counterpart algorithm, are trained on a series of training sets. These achieve a high performance area under the ROC curve of ≤0.97, outperforming go2ppi-a previously established prediction tool for protein-protein interactions (PPI) based on Gene Ontology (GO) annotations. Availability and implementation: https://github.com/ima23/maxent-ppi. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Sub-cellular localisation of proteins is an essential post-translational regulatory mechanism that can be assayed using high-throughput mass spectrometry (MS). These MS-based spatial proteomics experiments enable us to pinpoint the sub-cellular distribution of thousands of proteins in a specific system under controlled conditions. Recent advances in high-throughput MS methods have yielded a plethora of experimental spatial proteomics data for the cell biology community. Yet, there are many third-party data sources, such as immunofluorescence microscopy or protein annotations and sequences, which represent a rich and vast source of complementary information. We present a unique transfer learning classification framework that utilises a nearest-neighbour or support vector machine system, to integrate heterogeneous data sources to considerably improve on the quantity and quality of sub-cellular protein assignment. We demonstrate the utility of our algorithms through evaluation of five experimental datasets, from four different species in conjunction with four different auxiliary data sources to classify proteins to tens of sub-cellular compartments with high generalisation accuracy. We further apply the method to an experiment on pluripotent mouse embryonic stem cells to classify a set of previously unknown proteins, and validate our findings against a recent high resolution map of the mouse stem cell proteome. The methodology is distributed as part of the open-source Bioconductor pRoloc suite for spatial proteomics data analysis.

Smartphone sensing research has been advancing at a brisk pace. Yet, current social networking services often only take advantage of location sensing: applications like Foursquare use the phone's GPS and Wi-Fi radios to infer the user's location to simplify checking-in to a place. However, smartphone sensing could be exploited to considerably expand the spectrum of information a user can share with a few clicks with friends: not only the location of an event but activities such as “cooking dinner” or “waiting for a bus” can be predicted and suggested to the user to ease the check-in process. In this paper we show how mobile phone sensing can be used in this sense. For this prediction process to be accurate however, sensors need to be sampled often, with a considerable impact on the phone battery. To alleviate this issue, we explore streams of phone usage data (soft sensors), such as application usage, messages, and phone calls for predicting the user's activity in a more efficient fashion for augmenting mobile social check-in apps. We have deployed our application and collected a dataset of over 2700 check-ins to 48 activities from 20 users. Our analysis shows a prediction accuracy of 75% when offering 5 check-in suggestions to users. Furthermore, we show that when using only soft sensors we can achieve very similar performance to that obtained with real sensors, thereby significantly reducing the impact on the phone battery. This finding might have a potentially high impact on smartphone based activity check-in apps.

We propose a Gaussian process (GP) framework for robust inference in which a GP prior on the mixing weights of a two-component noise model augments the standard process over latent function values. This approach is a generalization of the mixture likelihood used in traditional robust GP regression, and a specialization of the GP mixture models suggested by Tresp (2000) and Rasmussen and Ghahramani (2002). The value of this restriction is in its tractable expectation propagation updates, which allow for faster inference and model selection, and better convergence than the standard mixture. An additional benefit over the latter method lies in our ability to incorporate knowledge of the noise domain to influence predictions, and to recover with the predictive distribution information about the outlier distribution via the gating process. The model has asymptotic complexity equal to that of conventional robust methods, but yields more confident predictions on benchmark problems than classical heavy-tailed models and exhibits improved stability for data with clustered corruptions, for which they fail altogether. We show further how our approach can be used without adjustment for more smoothly heteroscedastic data, and suggest how it could be extended to more general noise models. We also address similarities with the work of Goldberg et al. (1998), and the more recent contributions of Tresp, and Rasmussen and Ghahramani.

We present an efficient generalization of the sparse pseudo-input Gaussian pro- cess (SPGP) model developed by Snelson and Ghahramani [1], applying it to binary classification problems. By taking advantage of the SPGP prior covari- ance structure, we derive a numerically stable algorithm with O(N M 2) training complexity—asymptotically the same as related sparse methods such as the in- formative vector machine [2], but which more faithfully represents the posterior. We present experimental results for several benchmark problems showing that in many cases this allows an exceptional degree of sparsity without compromis- ing accuracy. Following [1], we locate pseudo-inputs by gradient ascent on the marginal likelihood, but exhibit occasions when this is likely to fail, for which we suggest alternative solutions.

Article Cross-validation for binary classification by real-valued functions: theoretical analysis Share on Authors: Martin Anthony Department of Mathematics, London School of Economics, Houghton Street, London WC2A 2AE, U.K. Department of Mathematics, London School of Economics, Houghton Street, London WC2A 2AE, U.K.View Profile , Sean B. Holden Department of Computer Science, University College London, Gower Street, London WC1E 6BT, U.K. Department of Computer Science, University College London, Gower Street, London WC1E 6BT, U.K.View Profile Authors Info & Claims COLT' 98: Proceedings of the eleventh annual conference on Computational learning theoryJuly 1998 Pages 218–229https://doi.org/10.1145/279943.279987Online:24 July 1998Publication History 10citation449DownloadsMetricsTotal Citations10Total Downloads449Last 12 Months29Last 6 weeks5 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

The application of statistical physics to the study of the learning curves of feedforward connectionist networks has to date been concerned mostly with perceptron-like networks. Recent work has extended the theory to networks such as committee machines and parity machines, and an important direction for current and future research is the extension of this body of theory to further connectionist networks. In this article, we use this formalism to investigate the learning curves of gaussian radial basis function networks (RBFNs) having fixed basis functions. (These networks have also been called generalized linear regression models.) We address the problem of learning linear and nonlinear, realizable and unrealizable, target rules from noise-free training examples using a stochastic training algorithm. Expressions for the generalization error, defined as the expected error for a network with a given set of parameters, are derived for general gaussian RBFNs, for which all parameters, including centers and spread parameters, are adaptable. Specializing to the case of RBFNs with fixed basis functions (basis functions having parameters chosen without reference to the training examples), we then study the learning curves for these networks in the limit of high temperature.

When designing a pattern classifier it is often the case that we have available a supervised learning technique and a collection of training data, and we would like to gain some idea of what the error probability of our classifier will be after training.A popular way of approaching this problem in practice is to use some form of error estimate, one of the most popular estimates being the cross-validation estimate.In this paper we address the following question: if we have n training examples what is the probability that the leave-one-out cross-validation estimate differs from the actual error probability by more than a constant c?We derive upper bounds on this probability for the closure algorithm and the deterministic l-inclusion graph prediction strategy.1

This article addresses the question of whether some recent Vapnik-Chervonenkis (VC) dimension-based bounds on sample complexity can be regarded as a practical design tool. Specifically, we are interested in bounds on the sample complexity for the problem of training a pattern classifier such that we can expect it to perform valid generalization. Early results using the VC dimension, while being extremely powerful, suffered from the fact that their sample complexity predictions were rather impractical. More recent results have begun to improve the situation by attempting to take specific account of the precise algorithm used to train the classifier. We perform a series of experiments based on a task involving the classification of sets of vowel formant frequencies. The results of these experiments indicate that the more recent theories provide sample complexity predictions that are significantly more applicable in practice than those provided by earlier theories; however, we also find that the recent theories still have significant shortcomings.

The ability of connectionist networks to generalize is often cited as one of their most important properties. We analyze the generalization ability of the class of generalized single-layer networks (GSLNs), which includes Volterra networks, radial basis function networks, regularization networks, and the modified Kanerva model, using techniques based on the theory of probably approximately correct (PAC) learning which have previously been used to analyze the generalization ability of feedforward networks of linear threshold elements (LTEs). An introduction to the relevant computational learning theory is included. We derive necessary and sufficient conditions on the number of training examples required by a GSLN to guarantee a particular generalization performance. We compare our results to those given previously for feedforward networks of LTEs and show that, on the basis of the currently available bounds, the sufficient number of training examples for GSLNs is typically considerably less than for feedforward networks of LTEs with the same number of weights. We show that the use of self-structuring techniques for GSLNs may reduce the number of training examples sufficient to guarantee good generalization performance, and we provide an explanation for the fact that GSLNs can require a relatively large number of weights.

In this paper we examine the representational and expressive power of two types of linearly weighted neural network:the polynomial discriminators (PDFs) and the radial basis function networks (RBFNs)..

The pattern classification and generalization ability of the class of generalized single-layer networks (GSLNs) using techniques from computational learning theory is analyzed. The authors derive necessary and sufficient conditions on the number of training examples required in order to guarantee a particular generalization performance and compare the bounds obtained to those available for (multilayer) feedforward networks of linear threshold elements (LTEs). This allows one to show that, on the basis of currently available bounds, the sufficient number of training examples for GSLNs tends to be considerably less than for feedforward networks of LTEs with the same number of weights. It is also shown that the use of self-structuring techniques for GSLNs may reduce the number of training examples sufficient for good generalization. An explanation for the fact that GSLNs can require a relatively large number of weights is given.>

The authors describe the development of a new form of connectionist model or neural network which, although having a unimodal error surface, will optimize its structure adaptively. The network is derived from the Volterra connectionist model. The adaptive noninformative extension nonlinearities removal method is used. The method requires only a slight computational expenditure in order to perform the self-structuring operation and may be operated with a bias to network contraction or reduction of error. Also, it does not require exact redundancies in order to operate. In addition, the inherent null-space nonlinearities removal method demonstrates a much more fundamental basis for removal of redundancy. This approach provides a deeper understanding of the problem and may in its current form be used as a block method with training set data.>