LearningFromSheep – Joint Chair

The joint research project LearningFromSheep copes with the use of advanced deep learning models to develop an innovative system for counting differential somatic cells—an innovative proxy for assessing milk quality and udder health—and to automate the analysis of microscope images, thereby eliminating the need for manual counting and drastically reducing data processing times.

FREECLIMB - Fruit Crops Resilience to Climate Change in the Mediterranean Basin

RABoLa - Sustainable strategies to reduce the use of antibiotics in dairy farming

The project, which involves different departments of the University of Milan (DISAA, DIMEVET, VESPA), researchers from the CNR ISPA and CNR IBBA and the DIANA Department of the Catholic University of the Sacred Heart of Piacenza, aims the primary identification of operative practices for reducing the use of antibiotics in the dairy cow herds. During the study, operative protocols already preliminary tested (administration of Aloe arborescens, use of Lactococcus lactis subsp. cremoris in pre and post dipping, molecules to inhibit Prototheca spp) will be validated during the entire lactation cycle of the cows.

This project will allow breeders to take advantages of new nutraceutical strategies and alternatives to the use of antibiotics, to enhance the innate defences and reduce the incidence and severity of mastitis. The validation of an effective protocol for the selective drying of animals would lead to a reduction of the preventive use of antibiotics, in line with what is requested at European level.

PNRR 2022 POLYPLOIDBREEDING 4.0 - Expanding the toolbox for cereal breeding: high-throughput genomics, 2D-3D phenomics and artificial intelligence for breeding with increasing genome complexity, from barley to durum and bread wheat

POLYPLOIDBREEDING 4.0 will focus on technology-driven breeding methods for the genetic improvement of crops (breeding 4.0) in diploid (barley, Hordeum vulgare) and polyploid (durum and bread wheat, Triticum durum, aestivum) species to increase cereal production in a sustainable and climate-smart way.

Target technologies are high throughput phenotyping (e.g. drone phenotyping, root scans from rhizotrons) and genotyping (e.g. SNP array, exome sequencing, GBS) for artificial intelligence based breeding. Phenotypes will include yield, morphometric measurements, UAV (unmanned aerial vehicle: drone)-captured image data linked to morphology and production efficiency and rhizotron-based root scans (both 2D and 3D). Machine-learning methods, focussing especially on deep learning methods, will be used for phenomic and whole-genome predictions of target phenotypes.

Sheep-TreeSeq: Scalable analysis of sheep genome diversity using tree sequences

Sheep-TreeSeq will perform scalable analysis of genomic diversity of sheep global populations using the novel tree sequence data format and methodology. Technological advances in agritech have increased the availability of genomic data, leading to massive datasets (“big data”) which pose challenges for storage, processing and analysis, e.g. the sheer volume of the data, the rapid generation of new data (updating results, expanding training populations, streaming applications), and the heterogeneity of data sources (integration of data from multiple sequencing and genotyping platforms). The tree sequence algorithm offers an excellent way to address such challenges, by providing lossless compression and novel representation of the data. As an example, using tree sequences on the 1000 Bull Genome Project data a 90% lossless compression was obtained, reducing the data size from ~800 GB to 45 GB. For the Sheep TreeSeq project we will use around 3,500 sheep whole genome sequences and over 50,000 genotypes (~10 TB of data). Our plan is to apply the tree sequence approach to compress the data and obtain a data representation highly suited for population genetics and demographic analysis: (i) principal component and genealogical nearest neighbour clustering; (ii) fixation index measuring genetic differentiation; (iii) deep neural network based clustering methods; iv) detection of runs of homozygosity (ROH) and heterozygosity-rich regions (HRR). This is the first time that this approach is applied to sheep genomics.

PRIN AGRITECH (SPOKE 1) - Plant, animal and microbial genetic resources and adaptation to climatic changes

Plant, animal and microbial genetic resources and adaptation to climatic changes

The resilience of agricultural and forest ecosystems under stressful conditions generated by climate change requires the enhancement and exploitation of genetic resources through state-of-the-art conservation strategies, combined with in-depth genome characterization and high-throughput phenotyping.

Activities include large-scale sequencing of accessions/breeds/strains, extensive marker-based genome descriptions, elucidation of the (pan)genome, deep phenotyping, and multi-omics characterization.

The resulting information, processed through advanced methods for the analysis, interpretation, archiving, and management of complex data, will highlight superior alleles/haplotypes, identify beneficial interactions across a variety of conditions, and define conservation units.

DeepMicroCore: Harnessing the power of artificial intelligence to identify the core microbiome

In DeepMicroCore we will develop DNN models to predict the substrate (species, tissue, food), time (e.g. developmental/physiological stage, experimental timepoint) and place (geographical location) of origin from microbiome data. We refer (and will use) specifically to 16S/18S/ITS rRNA-gene/amplicon sequencing data (metataxonomics). These predictive models will be optimized to maximise the accuracy of predictions in an unbiased way through cross-validation. With these working models, we will then be able to estimate variable importance and retrieve interesting features, e.g. the most relevant ASV/OTU for their impact on the prediction accuracy, or combinations and functions of ASV/OTU, or between-sample distances from embeddings. The features extracted from the DNN predictive models will then help us define and identify the core microbiome. For example, given microbiome data from multiple human/animal tissues (e.g. gut, skin, milk) sampled in different geographical locations, we will build a DNN model able to accurately predict to which tissue and location each sample belongs. Then, by extracting the relevant features from this model we will be able to identify the gut/skin/milk core microbiome. The core microbiome will be composed of those microbial taxa that discriminate between different tissues: some taxa may overlap between different microbiomes, and these will be identified by specific predictive models (e.g. one vs all predictions).

PROB-AI: Identifying novel probiotics from sequence data by means of artificial intelligence algorithms

The objective of the PROB-AI project is to train artificial intellgence (machine learning: ML) on a reduced set of confirmed probiotic species to then be able to identify new potential probiotics from all bacterial sequences available in public databases or research projects. The results of PROB-AI will help scientists to evaluate the specific features of probiotic and non-ptobiotic bacterial species, which will facilitate the discovery of new probiotics to be used in the biomedical and pharmaceutical sectors. Furthermore, PROB-AI will validate experimentally the new probiotics detected in silico by the algortihms, by using in vitro and in vivo validation approaches, and by highlighting the fundamental characteristics that tell apart symbiotic from pathogenic microorganisms in the human digestive tract. The use of ML for the accurate identification of novel probiotics is a highly innovative approach, still underexplored in the field of microbiology.

STM (Short-term Mobility) - To impute or not impute, a problem of missing data

This STM (Short-term Mobility) 2025 project is a collaboration with the School of Biosciences, Cardiff University, to explore the efficiency of imputing missing genotyping data in different species under case-scenarios involving relatively small sample sizes, large proportions of missing data and heterogeneous populations.

The genomic revolution increased the quantity of genetic data that can be generated, but the problem of missing data skewing statistical analyses has remained an issue. The main objective of this project is to determine the reliability of imputation approaches used to fill in missing genotypes in genomic datasets. This has been achieved using bioinformatic approaches to i) quantify the accuracy of imputation and ii) quantify the sensitivity of imputation to different levels of genetic divergence.

PRIN 2022 MISS.GO – Microbiota and Immune SignatureS of Graves’ disease and Orbitopathy

MISSGO addresses Graves’ disease and its main complication, Graves’ orbitopathy, an inflammatory condition that can cause severe functional and aesthetic impairment. As predictive models for identifying patients at risk of developing orbitopathy are still lacking, the project aims to clarify the underlying biological mechanisms.

MISSGO investigates the interaction between the gut microbiota and the immune system, with a particular focus on the imbalance between pro-inflammatory and regulatory T cells. Using advanced shotgun sequencing techniques integrated with multi-omics approaches, the project analyses microbiota composition and immune responses in patients with Graves’ disease, Graves’ orbitopathy, and healthy controls.

A key innovative aspect is the study of immune cells directly sampled from the thyroid and lymph nodes, enabling a disease-specific characterization. By integrating immunological and microbiological data, MISSGO aims to develop a predictive model for early identification of patients at risk, opening new perspectives for timely diagnosis and targeted therapies

FARM-INN Farm-level interventions supporting dairy industry innovation

The project FARM-INN aims to provide farm-level interventions supporting dairy industry enhancing safety and quality of milk and cheese and providing the necessary scientific evidence and new insight regarding their functional properties. The proposed actions will be carried out assessing and improving animal welfare and the environmental sustainability. In particular, two aspects will be tackled and studied in the proposed project: i) the development and use of new feed supplements adsorbing mycotoxins and reducing pathogenic and spoilage clostridia in milk ii) the characterization of cheese making and functional properties of A1 and A2 variants of beta-casein in milk. The evaluation of environmental sustainability of the adsorbent supplementations to the cow rations and the potential effect of cheese making using A1 and A2 beta-casein types in milk analyses through a Life Cycle Assessment approach will be performed.

The project will offer opportunities to the dairy farmers to strengthen their competitiveness, in the context of a better control of safety and quality issues, and will help dairy industry in placing on the market high-quality products adapted to the new expectations of consumers.

USEFUL - Development of techniques and supply chain processes (farms and dairies) to optimize environmental, territorial and managerial factors aimed at obtaining greater production efficiency and qualitative excellence in the production of DOP and typical cheeses

The USEFUL project aims to solve some critical issues in the production of Lombardy PDO cheeses (Grana Padano and Taleggio) in order to guarantee the competitiveness of the agro-industry companies. To this end, supply chain strategies and operational tools will be developed that allow to improve the microbiological and cheese-making quality of the milk and to improve all the cheese-making processes. These objectives will be achieved through the integration of the 3 sub-projects: Coordination, Innovation and Technological Transfer.

The expected results are: 1) to identify new procedures to improve animal health, food safety and the microbiological quality and cheese-making properties of milk; 2) to identify innovative procedures and operational tools to be used in the dairy industry to monitor the cheese-making and maturing phases of the cheese; 3) to define guidelines, technical manuals and fact sheets on the various procedures and techniques.