Mario Marchand

Domain-adversarial training of neural networks

We introduce a new representation learning approach for domain adaptation, in which data at training and test time come from similar but different distributions. Our approach is directly inspired by the theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predictions must be made based on features that cannot discriminate between the training (source) and test (target) domains. The approach implements this idea in the context of neural network architectures that are trained on labeled data from the source domain and unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of features that are (i) discriminative for the main learning task on the source domain and (ii) indiscriminate with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation and stochastic gradient descent, and can thus be implemented with little effort using any of the deep learning packages. We demonstrate the success of our approach for two distinct classification problems (document sentiment analysis and image classification), where state-of-the-art domain adaptation performance on standard benchmarks is achieved. We also validate the approach for descriptor learning task in the context of person re-identification application.

A convergence theorem for sequential learning in two-layer perceptrons

We consider a perceptron with Ni input units, one output and a yet unspecified number of hidden units. This perceptron must be able to learn a given but arbitrary set of input-output examples. By sequential learning we mean that groups of patterns, pertaining to the same class, are sequentially separated from the rest by successively adding hidden units until the remaining patterns are all in the same class. We prove that the internal representations obtained by such procedures are linearly separable. Preliminary numerical tests of an algorithm implementing these ideas are presented and compare favourably with results of other growth algorithms.

PAC-Bayesian learning of linear classifiers

We present a general PAC-Bayes theorem from which all known PAC-Bayes risk bounds are obtained as particular cases. We also propose different learning algorithms for finding linear classifiers that minimize these bounds. These learning algorithms are generally competitive with both AdaBoost and the SVM.

A Growth Algorithm for Neural Network Decision Trees

This paper explores the application of neural network principles to the construction of decision trees from examples. We consider the problem of constructing a tree of perceptrons able to execute a given but arbitrary Boolean function defined on Ni input bits. We apply a sequential (from one tree level to the next) and parallel (for neurons in the same level) learning procedure to add hidden units until the task in hand is performed. At each step, we use a perceptron-type algorithm over a suitable defined input space to minimize a classification error. The internal representations obtained in this way are linearly separable. Preliminary results of this algorithm are presented.

Feature Selection with Conjunctions of Decision Stumps and Learning from Microarray Data

One of the objectives of designing feature selection learning algorithms is to obtain classifiers that depend on a small number of attributes and have verifiable future performance guarantees. There are few, if any, approaches that successfully address the two goals simultaneously. To the best of our knowledge, such algorithms that give theoretical bounds on the future performance have not been proposed so far in the context of the classification of gene expression data. In this work, we investigate the premise of learning a conjunction (or disjunction) of decision stumps in Occams Razor, Sample Compression, and PAC-Bayes learning settings for identifying a small subset of attributes that can be used to perform reliable classification tasks. We apply the proposed approaches for gene identification from DNA microarray data and compare our results to those of the well-known successful approaches proposed for the task. We show that our algorithm not only finds hypotheses with a much smaller number of genes while giving competitive classification accuracy but also having tight risk guarantees on future performance, unlike other approaches. The proposed approaches are general and extensible in terms of both designing novel algorithms and application to other domains.

HIV-1 coreceptor usage prediction without multiple alignments: an application of string kernels

BackgroundHuman immunodeficiency virus type 1 (HIV-1) infects cells by means of ligand-receptor interactions. This lentivirus uses the CD4 receptor in conjunction with a chemokine coreceptor, either CXCR4 or CCR5, to enter a target cell. HIV-1 is characterized by high sequence variability. Nonetheless, within this extensive variability, certain features must be conserved to define functions and phenotypes. The determination of coreceptor usage of HIV-1, from its protein envelope sequence, falls into a well-studied machine learning problem known as classification. The support vector machine (SVM), with string kernels, has proven to be very efficient for dealing with a wide class of classification problems ranging from text categorization to protein homology detection. In this paper, we investigate how the SVM can predict HIV-1 coreceptor usage when it is equipped with an appropriate string kernel.ResultsThree string kernels were compared. Accuracies of 96.35% (CCR5) 94.80% (CXCR4) and 95.15% (CCR5 and CXCR4) were achieved with the SVM equipped with the distant segments kernel on a test set of 1425 examples with a classifier built on a training set of 1425 examples. Our datasets are built with Los Alamos National Laboratory HIV Databases sequences. A web server is available at http://genome.ulaval.ca/hiv-dskernel.ConclusionWe examined string kernels that have been used successfully for protein homology detection and propose a new one that we call the distant segments kernel. We also show how to extract the most relevant features for HIV-1 coreceptor usage. The SVM with the distant segments kernel is currently the best method described.

Learning a peptide-protein binding affinity predictor with kernel ridge regression

BackgroundThe cellular function of a vast majority of proteins is performed through physical interactions with other biomolecules, which, most of the time, are other proteins. Peptides represent templates of choice for mimicking a secondary structure in order to modulate protein-protein interaction. They are thus an interesting class of therapeutics since they also display strong activity, high selectivity, low toxicity and few drug-drug interactions. Furthermore, predicting peptides that would bind to a specific MHC alleles would be of tremendous benefit to improve vaccine based therapy and possibly generate antibodies with greater affinity. Modern computational methods have the potential to accelerate and lower the cost of drug and vaccine discovery by selecting potential compounds for testing in silico prior to biological validation.ResultsWe propose a specialized string kernel for small bio-molecules, peptides and pseudo-sequences of binding interfaces. The kernel incorporates physico-chemical properties of amino acids and elegantly generalizes eight kernels, comprised of the Oligo, the Weighted Degree, the Blended Spectrum, and the Radial Basis Function. We provide a low complexity dynamic programming algorithm for the exact computation of the kernel and a linear time algorithm for it’s approximation. Combined with kernel ridge regression and SupCK, a novel binding pocket kernel, the proposed kernel yields biologically relevant and good prediction accuracy on the PepX database. For the first time, a machine learning predictor is capable of predicting the binding affinity of any peptide to any protein with reasonable accuracy. The method was also applied to both single-target and pan-specific Major Histocompatibility Complex class II benchmark datasets and three Quantitative Structure Affinity Model benchmark datasets.ConclusionOn all benchmarks, our method significantly (p-value ≤ 0.057) outperforms the current state-of-the-art methods at predicting peptide-protein binding affinities. The proposed approach is flexible and can be applied to predict any quantitative biological activity. Moreover, generating reliable peptide-protein binding affinities will also improve system biology modelling of interaction pathways. Lastly, the method should be of value to a large segment of the research community with the potential to accelerate the discovery of peptide-based drugs and facilitate vaccine development. The proposed kernel is freely available at http://graal.ift.ulaval.ca/downloads/gs-kernel/.

Learning Nonoverlapping Perceptron Networks from Examples and Membership Queries

We investigate, within the PAC learning model, the problem of learning nonoverlapping perceptron networks (also known as read-once formulas over a weighted threshold basis). These are loop-free neural nets in which each node has only one outgoing weight. We give a polynomial time algorithm that PAC learns any nonoverlapping perceptron network using examples and membership queries. The algorithm is able to identify both the architecture and the weight values necessary to represent the function to be learned. Our results shed some light on the effect of the overlap on the complexity of learning in neural networks.

Machine learning assisted design of highly active peptides for drug discovery

The discovery of peptides possessing high biological activity is very challenging due to the enormous diversity for which only a minority have the desired properties. To lower cost and reduce the time to obtain promising peptides, machine learning approaches can greatly assist in the process and even partly replace expensive laboratory experiments by learning a predictor with existing data or with a smaller amount of data generation. Unfortunately, once the model is learned, selecting peptides having the greatest predicted bioactivity often requires a prohibitive amount of computational time. For this combinatorial problem, heuristics and stochastic optimization methods are not guaranteed to find adequate solutions. We focused on recent advances in kernel methods and machine learning to learn a predictive model with proven success. For this type of model, we propose an efficient algorithm based on graph theory, that is guaranteed to find the peptides for which the model predicts maximal bioactivity. We also present a second algorithm capable of sorting the peptides of maximal bioactivity. Extensive analyses demonstrate how these algorithms can be part of an iterative combinatorial chemistry procedure to speed up the discovery and the validation of peptide leads. Moreover, the proposed approach does not require the use of known ligands for the target protein since it can leverage recent multi-target machine learning predictors where ligands for similar targets can serve as initial training data. Finally, we validated the proposed approach in vitro with the discovery of new cationic antimicrobial peptides. Source code freely available at http://graal.ift.ulaval.ca/peptide-design/.

MHC-NP: Predicting peptides naturally processed by the MHC

We present MHC-NP, a tool for predicting peptides naturally processed by the MHC pathway. The method was part of the 2nd Machine Learning Competition in Immunology and yielded state-of-the-art accuracy for the prediction of peptides eluted from human HLA-A*02:01, HLA-B*07:02, HLA-B*35:01, HLA-B*44:03, HLA-B*53:01, HLA-B*57:01 and mouse H2-D(b) and H2-K(b) MHC molecules. We briefly explain the theory and motivations that have led to developing this tool. General applicability in the field of immunology and specifically epitope-based vaccine are expected. Our tool is freely available online and hosted by the Immune Epitope Database at http://tools.immuneepitope.org/mhcnp/.