The Immunopeptidome from a Genomic Perspective: Establishing the Noncanonical Landscape of MHC Class I-Associated Peptides.

Tumor antigens can emerge through multiple mechanisms, including translation of noncoding genomic regions. This noncanonical category of tumor antigens has recently gained attention; however, our understanding of how they recur within and between cancer types is still in its infancy. Therefore, we developed a proteogenomic pipeline based on deep learning de novo mass spectrometry (MS) to enable the discovery of noncanonical MHC class I-associated peptides (ncMAP) from noncoding regions. Considering that the emergence of tumor antigens can also involve posttranslational modifications (PTM), we included an open search component in our pipeline. Leveraging the wealth of MS-based immunopeptidomics, we analyzed data from 26 MHC class I immunopeptidomic studies across 11 different cancer types. We validated the de novo identified ncMAPs, along with the most abundant PTMs, using spectral matching and controlled their FDR to 1%. The noncanonical presentation appeared to be 5 times enriched for the A03 HLA supertype, with a projected population coverage of 55%. The data reveal an atlas of 8,601 ncMAPs with varying levels of cancer selectivity and suggest 17 cancer-selective ncMAPs as attractive therapeutic targets according to a stringent cutoff. In summary, the combination of the open-source pipeline and the atlas of ncMAPs reported herein could facilitate the identification and screening of ncMAPs as targets for T-cell therapies or vaccine development.

Syntax and prejudice: ethically-charged biases of a syntax-based hate speech recognizer unveiled.

Hate speech recognizers (HSRs) can be the panacea for containing hate in social media or can result in the biggest form of prejudice-based censorship hindering people to express their true selves. In this paper, we hypothesized how massive use of syntax can reduce the prejudice effect in HSRs. To explore this hypothesis, we propose Unintended-bias Visualizer based on Kermit modeling (KERM-HATE): a syntax-based HSR, which is endowed with syntax heat parse trees used as a post-hoc explanation of classifications. KERM-HATE significantly outperforms BERT-based, RoBERTa-based and XLNet-based HSR on standard datasets. Surprisingly this result is not sufficient. In fact, the post-hoc analysis on novel datasets on recent divisive topics shows that even KERM-HATE carries the prejudice distilled from the initial corpus. Therefore, although tests on standard datasets may show higher performance, syntax alone cannot drive the "attention" of HSRs to ethically-unbiased features.

Data Augmentation Using Background Replacement for Automated Sorting of Littered Waste.

The introduction of sophisticated waste treatment plants is making the process of trash sorting and recycling more and more effective and eco-friendly. Studies on Automated Waste Sorting (AWS) are greatly contributing to making the whole recycling process more efficient. However, a relevant issue, which remains unsolved, is how to deal with the large amount of waste that is littered in the environment instead of being collected properly. In this paper, we introduce BackRep: a method for building waste recognizers that can be used for identifying and sorting littered waste directly where it is found. BackRep consists of a data-augmentation procedure, which expands existing datasets by cropping solid waste in images taken on a uniform (white) background and superimposing it on more realistic backgrounds. For our purpose, realistic backgrounds are those representing places where solid waste is usually littered. To experiment with our data-augmentation procedure, we produced a new dataset in realistic settings. We observed that waste recognizers trained on augmented data actually outperform those trained on existing datasets. Hence, our data-augmentation procedure seems a viable approach to support the development of waste recognizers for urban and wild environments.

Evaluating diagnostic content of AI-generated radiology reports of chest X-rays.

Radiology reports are of core importance for the communication between the radiologist and clinician. A computer-aided radiology report system can assist radiologists in this task and reduce variation between reports thus facilitating communication with the medical doctor or clinician. Producing a well structured, clear, and clinically well-focused radiology report is essential for high-quality patient diagnosis and care. Despite recent advances in deep learning for image caption generation, this task remains highly challenging in a medical setting. Research has mainly focused on the design of tailored machine learning methods for this task, while little attention has been devoted to the development of evaluation metrics to assess the quality of AI-generated documents. Conventional quality metrics for natural language processing methods like the popular BLEU score, provide little information about the quality of the diagnostic content of AI-generated radiology reports. In particular, because radiology reports often use standardized sentences, BLEU scores of generated reports can be high while they lack diagnostically important information. We investigate this problem and propose a new measure that quantifies the diagnostic content of AI-generated radiology reports. In addition, we exploit the standardization of reports by generating a sequence of sentences. That is, instead of using a dictionary of words, as current image captioning methods do, we use a dictionary of sentences. The assumption underlying this choice is that radiologists use a well-focused vocabulary of 'standard' sentences, which should suffice for composing most reports. As a by-product, a significant training speed-up is achieved compared to models trained on a dictionary of words. Overall, results of our investigation indicate that standard validation metrics for AI-generated documents are weakly correlated with the diagnostic content of the reports. Therefore, these measures should be not used as only validation metrics, and measures evaluating diagnostic content should be preferred in such a medical context.

Towards the Interpretability of Machine Learning Predictions for Medical Applications Targeting Personalised Therapies: A Cancer Case Survey.

Artificial Intelligence is providing astonishing results, with medicine being one of its favourite playgrounds. Machine Learning and, in particular, Deep Neural Networks are behind this revolution. Among the most challenging targets of interest in medicine are cancer diagnosis and therapies but, to start this revolution, software tools need to be adapted to cover the new requirements. In this sense, learning tools are becoming a commodity but, to be able to assist doctors on a daily basis, it is essential to fully understand how models can be interpreted. In this survey, we analyse current machine learning models and other in-silico tools as applied to medicine-specifically, to cancer research-and we discuss their interpretability, performance and the input data they are fed with. Artificial neural networks (ANN), logistic regression (LR) and support vector machines (SVM) have been observed to be the preferred models. In addition, convolutional neural networks (CNNs), supported by the rapid development of graphic processing units (GPUs) and high-performance computing (HPC) infrastructures, are gaining importance when image processing is feasible. However, the interpretability of machine learning predictions so that doctors can understand them, trust them and gain useful insights for the clinical practice is still rarely considered, which is a factor that needs to be improved to enhance doctors' predictive capacity and achieve individualised therapies in the near future.

Predicting VTE in Cancer Patients: Candidate Biomarkers and Risk Assessment Models.

Risk prediction of chemotherapy-associated venous thromboembolism (VTE) is a compelling challenge in contemporary oncology, as VTE may result in treatment delays, impaired quality of life, and increased mortality. Current guidelines do not recommend thromboprophylaxis for primary prevention, but assessment of the patient's individual risk of VTE prior to chemotherapy is generally advocated. In recent years, efforts have been devoted to building accurate predictive tools for VTE risk assessment in cancer patients. This review focuses on candidate biomarkers and prediction models currently under investigation, considering their advantages and disadvantages, and discussing their diagnostic performance and potential pitfalls.

Symbolic, Distributed, and Distributional Representations for Natural Language Processing in the Era of Deep Learning: A Survey.

Natural language is inherently a discrete symbolic representation of human knowledge. Recent advances in machine learning (ML) and in natural language processing (NLP) seem to contradict the above intuition: discrete symbols are fading away, erased by vectors or tensors called distributed and distributional representations. However, there is a strict link between distributed/distributional representations and discrete symbols, being the first an approximation of the second. A clearer understanding of the strict link between distributed/distributional representations and symbols may certainly lead to radically new deep learning networks. In this paper we make a survey that aims to renew the link between symbolic representations and distributed/distributional representations. This is the right time to revitalize the area of interpreting how discrete symbols are represented inside neural networks.

Risk Assessment for Venous Thromboembolism in Chemotherapy-Treated Ambulatory Cancer Patients.

OBJECTIVE: To design a precision medicine approach aimed at exploiting significant patterns in data, in order to produce venous thromboembolism (VTE) risk predictors for cancer outpatients that might be of advantage over the currently recommended model (Khorana score). DESIGN: Multiple kernel learning (MKL) based on support vector machines and random optimization (RO) models were used to produce VTE risk predictors (referred to as machine learning [ML]-RO) yielding the best classification performance over a training (3-fold cross-validation) and testing set. RESULTS: Attributes of the patient data set ( n = 1179) were clustered into 9 groups according to clinical significance. Our analysis produced 6 ML-RO models in the training set, which yielded better likelihood ratios (LRs) than baseline models. Of interest, the most significant LRs were observed in 2 ML-RO approaches not including the Khorana score (ML-RO-2: positive likelihood ratio [+LR] = 1.68, negative likelihood ratio [-LR] = 0.24; ML-RO-3: +LR = 1.64, -LR = 0.37). The enhanced performance of ML-RO approaches over the Khorana score was further confirmed by the analysis of the areas under the Precision-Recall curve (AUCPR), and the approaches were superior in the ML-RO approaches (best performances: ML-RO-2: AUCPR = 0.212; ML-RO-3-K: AUCPR = 0.146) compared with the Khorana score (AUCPR = 0.096). Of interest, the best-fitting model was ML-RO-2, in which blood lipids and body mass index/performance status retained the strongest weights, with a weaker association with tumor site/stage and drugs. CONCLUSIONS: Although the monocentric validation of the presented predictors might represent a limitation, these results demonstrate that a model based on MKL and RO may represent a novel methodological approach to derive VTE risk classifiers. Moreover, this study highlights the advantages of optimizing the relative importance of groups of clinical attributes in the selection of VTE risk predictors.