publications

2024

1. Robustness and reproducibility for AI learning in biomedical sciences: RENOIR

   Alessandro Barberis, Hugo J.W.L. Aerts, and Francesca M. Buffa

   Scientific Reports, Jan 2024

   Abs

   Artificial intelligence (AI) techniques are increasingly applied across various domains, favoured by the growing acquisition and public availability of large, complex datasets. Despite this trend, AI publications often suffer from lack of reproducibility and poor generalisation of findings, undermining scientific value and contributing to global research waste. To address these issues and focusing on the learning aspect of the AI field, we present RENOIR (REpeated random sampliNg fOr machIne leaRning), a modular open-source platform for robust and reproducible machine learning (ML) analysis. RENOIR adopts standardised pipelines for model training and testing, introducing elements of novelty, such as the dependence of the performance of the algorithm on the sample size. Additionally, RENOIR offers automated generation of transparent and usable reports, aiming to enhance the quality and reproducibility of AI studies. To demonstrate the versatility of our tool, we applied it to benchmark datasets from health, computer science, and STEM (Science, Technology, Engineering, and Mathematics) domains. Furthermore, we showcase RENOIR’s successful application in recently published studies, where it identified classifiers for SET2D and TP53 mutation status in cancer. Finally, we present a use case where RENOIR was employed to address a significant pharmacological challenge—predicting drug efficacy. RENOIR is freely available at https://github.com/alebarberis/renoir .

2. Antitumour effect of the mitochondrial complex III inhibitor Atovaquone in combination with anti-PD-L1 therapy in mouse cancer models

   Gonzalo Rodriguez-Berriguete, Rathi Puliyadi, Nicole Machado, and 6 more authors

   Cell Death & Disease, Jan 2024

   Abs

   Immune checkpoint blockade (ICB) provides effective and durable responses for several tumour types by unleashing an immune response directed against cancer cells. However, a substantial number of patients treated with ICB develop relapse or do not respond, which has been partly attributed to the immune-suppressive effect of tumour hypoxia. We have previously demonstrated that the mitochondrial complex III inhibitor atovaquone alleviates tumour hypoxia both in human xenografts and in cancer patients by decreasing oxygen consumption and consequently increasing oxygen availability in the tumour. Here, we show that atovaquone alleviates hypoxia and synergises with the ICB antibody anti-PD-L1, significantly improving the rates of tumour eradication in the syngeneic CT26 model of colorectal cancer. The synergistic effect between atovaquone and anti-PD-L1 relied on CD8+ T cells, resulted in the establishment of a tumour-specific memory immune response, and was not associated with any toxicity. We also tested atovaquone in combination with anti-PD-L1 in the LLC (lung) and MC38 (colorectal) cancer syngeneic models but, despite causing a considerable reduction in tumour hypoxia, atovaquone did not add any therapeutic benefit to ICB in these models. These results suggest that atovaquone has the potential to improve the outcomes of patients treated with ICB, but predictive biomarkers are required to identify individuals likely to benefit from this intervention.

2023

1. A machine learning and directed network optimization approach to uncover TP53 regulatory patterns

   Charalampos P. Triantafyllidis, Alessandro Barberis, Fiona Hartley, and 6 more authors

   iScience, Dec 2023

   Abs

   Summary

   TP53, the Guardian of the Genome, is the most frequently mutated gene in human cancers and the functional characterization of its regulation is fundamental. To address this we employ two strategies: machine learning to predict the mutation status of TP53 from transcriptomic data, and directed regulatory networks to reconstruct the effect of mutations on the transcipt levels of TP53 targets. Using data from established databases (Cancer Cell Line Encyclopedia, The Cancer Genome Atlas), machine learning could predict the mutation status, but not resolve different mutations. On the contrary, directed network optimization allowed to infer the TP53 regulatory profile across: (1) mutations, (2) irradiation in lung cancer, and (3) hypoxia in breast cancer, and we could observe differential regulatory profiles dictated by (1) mutation type, (2) deleterious consequences of the mutation, (3) known hotspots, (4) protein changes, (5) stress condition (irradiation/hypoxia). This is an important first step toward using regulatory networks for the characterization of the functional consequences of mutations, and could be extended to other perturbations, with implications for drug design and precision medicine.

2. A role for SETD2 loss in tumorigenesis through DNA methylation dysregulation

   Hira Javaid, Alessandro Barberis, Olga Chervova, and 7 more authors

   BMC Cancer, Dec 2023

   Abs

   SETD2-dependent H3 Lysine-36 trimethylation (H3K36me3) has been recently linked to the deposition of de-novo DNA methylation. SETD2 is frequently mutated in cancer, however, the functional impact of SETD2 loss and depletion on DNA methylation across cancer types and tumorigenesis is currently unknown. Here, we perform a pan-cancer analysis and show that both SETD2 mutation and reduced expression are associated with DNA methylation dysregulation across 21 out of the 24 cancer types tested. In renal cancer, these DNA methylation changes are associated with altered gene expression of oncogenes, tumour suppressors, and genes involved in neoplasm invasiveness, including TP53, FOXO1, and CDK4. This suggests a new role for SETD2 loss in tumorigenesis and cancer aggressiveness through DNA methylation dysregulation. Moreover, using a robust machine learning methodology, we develop and validate a 3-CpG methylation signature which is sufficient to predict SETD2 mutation status with high accuracy and correlates with patient prognosis.

3. The impact of survivorship bias in glioblastoma research

   Alessandro Barberis, Francesco Pasqualetti, Sofia Zanotti, and 11 more authors

   Critical Reviews in Oncology/Hematology, Aug 2023

   Abs

   Despite advances in the therapy of Central Nervous System (CNS) malignancies, treatment of glioblastoma (GB) poses significant challenges due to GB resistance and high recurrence rates following post-operative radio-chemotherapy. The majority of prognostic and predictive GB biomarkers are currently developed using tumour samples obtained through surgical interventions. However, the selection criteria adopted by different neurosurgeons to determine which cases are suitable for surgery make operated patients not representative of all GB cases. Particularly, geriatric and frail individuals are excluded from surgical consideration in some cancer centers. Such selection generates a survival (or selection) bias that introduces limitations, rendering the patients or data chosen for downstream analyses not representative of the entire community. In this review, we discuss the implication of survivorship bias on current and novel biomarkers for patient selection, stratification, therapy, and outcome analyses.

2020

1. Disruption of hypoxia-inducible fatty acid binding protein 7 induces beige fat-like differentiation and thermogenesis in breast cancer cells

   Masahiro Kawashima, Karim Bensaad, Christos E. Zois, and 10 more authors

   Cancer & Metabolism, Jul 2020

2019

1. Guidelines for using sigQC for systematic evaluation of gene signatures

   Andrew Dhawan, Alessandro Barberis, Wei-Chen Cheng, and 6 more authors

   Nature Protocols, Apr 2019

   Abs Bib

   With the increased use of next-generation sequencing generating large amounts of genomic data, gene expression signatures are becoming critically important tools for the interpretation of these data, and are poised to have a substantial effect on diagnosis, management, and prognosis for a number of diseases. It is becoming crucial to establish whether the expression patterns and statistical properties of sets of genes, or gene signatures, are conserved across independent datasets. Conversely, it is necessary to compare established signatures on the same dataset to better understand how they capture different clinical or biological characteristics. Here we describe how to use sigQC, a tool that enables a streamlined, systematic approach for the evaluation of previously obtained gene signatures across multiple gene expression datasets. We implemented sigQC in an R package, making it accessible to users who have knowledge of file input/output and matrix manipulation in R and a moderate grasp of core statistical principles. SigQC has been adopted in basic biology and translational studies, including, but not limited to, the evaluation of multiple gene signatures for potential clinical use as cancer biomarkers. This protocol uses a previously obtained signature for breast cancer metastasis as an example to illustrate the critical quality control steps involved in evaluating its expression, variability, and structure in breast tumor RNA-sequencing data, a different dataset from that in which the signature was originally derived. We demonstrate how the outputs created from sigQC can be used for the evaluation of gene signatures on large-scale gene expression datasets.

   @article{Dhawan2019,
     author = {Dhawan, Andrew and Barberis, Alessandro and Cheng, Wei-Chen and Domingo, Enric and West, Catharine and Maughan, Tim and Scott, Jacob G. and Harris, Adrian L. and Buffa, Francesca Meteora},
     doi = {10.1038/s41596-019-0136-8},
     issn = {1754-2189},
     journal = {Nature Protocols},
     keywords = {Biomarkers,Data processing,Gene expression,Transcriptomics},
     mendeley-groups = {R},
     month = apr,
     number = {5},
     pages = {1377 -- 1400},
     publisher = {Nature Publishing Group},
     title = {{Guidelines for using sigQC for systematic evaluation of gene signatures}},
     url = {http://www.nature.com/articles/s41596-019-0136-8},
     volume = {14},
     year = {2019}
   }

2017

1. Serum microRNA expression during neoadjuvant chemoradiation for rectal cancer.

   Stephanie Hui-Su Lim, Emilia Ip, Wei Chua, and 8 more authors

   Journal of Clinical Oncology, May 2017

   Abs

   e15081Background: Changes in microRNA (miRNA) expression during treatment for locally advanced rectal cancer (LARC) may provide insight into disease biology and potentially act as predictive biomar...

2016

1. SCF (Fbxl17) ubiquitylation of Sufu regulates Hedgehog signaling and medulloblastoma development

   Madalina Raducu, Ella Fung, Sébastien Serres, and 13 more authors

   The EMBO Journal, Jul 2016
2. Paracrine effect of GTP cyclohydrolase and angiopoietin-1 interaction in stromal fibroblasts on tumor Tie2 activation and breast cancer growth

   Liye Chen, Xin Zeng, Esther Kleibeuker, and 9 more authors

   Oncotarget, Jan 2016

2015

1. Gene Expression Signatures as Biomarkers of Tumour Hypoxia

   Alessandro Barberis, Benjamin H. L. Harris, Catharine M. L. West, and 1 more author

   Clinical Oncology, Jan 2015

2013

1. Real-Time Implementation of the Vertex Component Analysis Algorithm on GPUs

   Alessandro Barberis, Gianni Danese, Francesco Leporati, and 2 more authors

   IEEE Geoscience and Remote Sensing Letters, Mar 2013

   Abs

   In this letter, we present a new parallel implemen- tation of the vertex component analysis (VCA) algorithm for spectral unmixing of remotely sensed hyperspectral data on com- modity graphics processing units.We first developed a C serial ver- sion of the VCA algorithm and three parallel versions: one using NVIDIA’s Compute Unified Device Architecture (CUDA), another using CUDA basic linear algebra subroutines library CUBLAS, and the last using the CUDA linear algebra library CULA. Ex- perimental results, based on the analysis of hyperspectral images acquired by a variety of hyperspectral imaging sensors, show the effectiveness of our implementation, which satisfies the real-time constraints given by the data acquisition rate.