Current research projects and collaborations
P. falciparum ookinete development and maturation
Plasmodium falciparum is a deadly parasite responsible for most of the malaria cases worldwide. This parasite requires a mosquito vector in order to infect humans. Inside the mosquito, the parasite goes through several life stages that are subject to intense investigations. The ookinete stage is the focus of this project. Here I developed a mathematical model that recapitulates the known experimental kinetic steps of the parasite in this stage. The model produces predictions that are currently subject of experimental investigation. This project is an experimental and theoretical collaboration with the group of Elena A. Levashina at the Max Planck Institute for Infection Biology in Berlin. The experimental work is performed by Giulia Costa and Pablo Suárez Cortés in the Levashina group. The results of this project may help uncovering differences in the development of the parasite under different conditions that could be better exploited in the fight against the disease.
An autoencoder to compress RNAseq data
Different tissues in mosquitoes involved in the transmission of the malaria parasite may respond differently when they are challenged with the parasite. Understanding the differential expression of genes in different tissues and different mosquito strains is key to understand how the mosquito immune system responds to the infection by the parasite. RNAseq data is usually collected from these tissues but both its high dimensionality and the small sample sizes are a challenge for the analysis of this data. In one Master’s thesis project we are developing an autoencoder, which is a tool to compress the data while possibly keeping the non-linear correlations among the genes. The compressed data should then provide a more useful set of information for downstream cluster analysis. This project is manly computational and is curried out by Clemens Bodenstein under the joint supervision of Elena A. Levashina (Max Planck Institute for Infection Biology in Berlin), and myself. Galo Rivera from the group of Elena A. Levashina performs the experimental work and is involved in the supervision. Finding out which genes are responsible for the immune reaction in different tissues of the mosquito will help elucidating how the mosquito fights against the parasite.
Artificial neural networks to aid molecular dynamics simulations
Molecular dynamics (MD) simulations are quite time expensive. The correct choice of the parameters of the molecular model (also known as force field parameters) is essential in order to reproduce the experimental results. The knowledge of the correct values of these parameters can be useful both for the information that they deliver about the molecular interactions and for the use of the MD tools in order to simulate similar but yet unexplored configurations and molecules. The search for the correct force field parameters is usually also a laborious endeavor. In collaboration with Markus Miettinen and Hanne Antila in the theory department of the Max Planck Institute of Colloids and Interfaces, we have started to develop an artificial neural network that should speed-up this search. Following the steps of Ella Steins successful Master’s thesis (TU Berlin), co-supervised by Markus Miettinen, Hanne Antila and myself, we set up with Rodrigo Lopez (a student of the Potsdam University) to develop further the artificial neural network. Once finished, the network will be trained on hundreds of MD simulations. Its predictions will provide an intelligent choice of the force field parameters to try next, thus tremendously reducing the development time of the correct MD tool.
An artificial neural network to classify AFM single molecule force spectroscopy
Coiled-coil interacting molecules are a simple single molecule interaction system whose mechanical properties can be characterized by means of single molecule force spectroscopy. When the AFM cantilever is attached to one of the two molecules and is moved away, the molecules stretch and eventually their bond is ruptured. The force required to reach rupture can be fine-tuned, at least in principle, by modifying the chemical composition and the pulling geometry during the experiment. Since coiled-coil molecular interaction can be quite useful in anchoring cells in artificial scaffolds, an understanding of their mechanical properties as function of their chemical composition is fundamental for their application in living systems. The analysis of the rupture force curves is subject to intensive studies and is still a challenging matter. Several features like the appearance of a plateau in the force extension curve just before the rupture event are difficult to detect by naked eyes in hundreds of curves but very important for the characterization of the molecular system. Zeynep Atris, a PhD student in the group of Kerstin Blank at the Max Planck Institute of Colloids and Interfaces, is developing an artificial neural network that once trained would be able to recognize when a force extension curve presents a plateau or not. This development of the neural network is supervised by Kerstin Blank and myself. The experimental part of the work is performed by Zeynep Atris under the supervision of Kerstin Blank.
Modeling tumor progression in bones
Contrary to common wisdom, human bone is not just a piece of dull material that we should take care of only when it breaks. In fact, bone is a living tissue that even in adults gets continuously remodeled, often in response to mechanical stimuli. The cells embedded into the bone material and other cells surrounding the bone interact with each other and exchange information as to where the bone needs to be remodeled. When cancer cells invade the tumor tissue, they engage in a dynamical interaction with the native bone remodeling cells. This interaction may disrupt the balance between the native bone remodeling cells thus seriously compromising the bone structure. To uncover the details of the interaction between cancer cells and bone remodeling cells, and the group of Amaia Cipitria in the department of biomaterials at the Max Planck Institute of Colloids and Interfaces is leading an intensive experimental work to shed light into this matter with her student Sarah Young. To understand the phenomenon of the interaction between bone remodeling cells and cancer cells, Anna Dorothea Heller, a PhD student co-supervised by Amaia Cipitria and myself, is developing a mathematical model that exploits the information provided by the experimental work. The model - a hybrid between a cellular automaton and a spatial evolutionary game- aims at determining the interaction between the cells involved. The interaction parameters that will be obtained when validating the model prediction with the experimental data will hopefully reveal biological phenomena to investigate with further experiments.
Tumor growth as an evolutionary game
Tumor progression is a very complex process driven by the accumulation of mutations in the cancer cells and by the interaction of the cancer cells with the microenvironment. In this project we focus our attention to the latter interaction. The microenvironment is made of “normal” cells, that may be grouped in one or more cell types, and cancer cells that we see as a homogenous population. We model this cell-cell interaction in terms of an evolutionary game in which the proliferation rate of every cell type depends on the interaction with (and frequency of) the other populations. Therefore, one aspect of tumor progression is tumor growth, namely the growth of the proportion of cancer cell in a tissue. Giacomo Rossato, a physics Master student from the University of Trieste currently visiting my group, is developing evolutionary game models that recapitulate most of the known clinical phenomenology of tumor growth when only the interaction with the microenvironment is considered. Under the supervision of Amaia Cipitria from the biomaterials department of the Max Planck Institute of Colloids and Interfaces and myself, Giacomo is working on an original model that shows how cancer cells can permanently modify their microenvironment thus enhancing their chance of growing even after an apparently successful therapeutic intervention.
Ribosome drop-off in yeast cells
Ribosomes are the machines that are responsible for the synthesis of proteins in the cells. Thanks to ribosomes, the protein content of the cells can increase thus eventually having enough material for a successful cell division. Nevertheless, as molecular machines also ribosomes are error prone. Together with Davide Chiarugi, currently head of the bioinformatics unit at the Institute of Metabolic Science of the University of Cambridge, we are looking at a special type of error that we call drop-off: this error occurs when the ribosome stops protein synthesis prematurely without any specific signal that it should do so. The rate of this unspecific termination of synthesis is a key parameter in understanding what is called the processivity of the molecular machine. So far, in a collaboration project between Davide Chiarugi and myself only the drop-off rate of the bacterium Escherichia coli has been determined. Sherine Awad, a postdoctoral fellow in the group of Davide Chiarugi, has taken the mammoth challenge to look at the drop-off rate of the yeast S. cerevisiae, another well-known model organism, by analyzing riso-seq data available in the literature. Under the supervision of Davide Chiarugi and myself, Sherine Awad’s results will finally shed light into the different level of processivity in eukaryotic cells compared to prokaryotic and hopefully help us understand why eukaryotic organisms tend to have longer genes than bacteria. Sherine Awad’s data analysis tool will also help analyzing ribo-seq data of other organisms.
Uncovering slow and fast translated regions in E. coli transcriptome
When ribosome translate a messenger RNA they don’t do it at a uniform speed. This fact is well-known in molecular biology. Many studies looking into this phenomenon have tried to uncover both the mechanisms responsible for local slow-downs and its possible consequences for the expression of the genes. Together with Davide Chiarugi, currently head of the bioinformatics unit at the Institute of Metabolic Science of the University of Cambridge, we developed a method to look for the subsequences in the transcriptome of Escherichia coli where ribosomes are slow (or fast) in a reproducible manner when comparing many independent experimental studies under the same growth conditions. One big challenge in this analysis is the low signal-to-noise ratio, which requires ad-hoc tools to harvest a few sequences. Giorgia Giacomini, a biology PhD student, together with Paolo Andreini, a machine learning engineer, both from the group of professor Monica Bianchini at the University of Siena, have taken the data available and developed a program to dig into the data following the initial pilot study of Davide Chiarugi and myself. Under the supervision of Davide Chiarugi, Monica Bianchini and myself, their analysis tool will produce a list of highly reliable and reproducible sequences from which we hope to learn the patterns that make up a fast versus a slow translated region.
A machine learning approach to map the slow and fast regions in the transcriptome
Starting from preliminary results of Davide Chiarugi, currently head of the bioinformatics unit at the Institute of Metabolic Science of the University of Cambridge, and myself, the group of Monica Bianchini at the University of Siena is developing deep learning tools to correctly classify those sub-sequences in the transcriptome of Escherichia coli where the ribosomes are slow or fast. This preliminary study, carried on by Pietro Bongini, Veronica Lachi and Caterina Graziani, PhD students in the lab of Monica Bianchini, has shown that these subsequences can be classified with a validation accuracy of above 80%, thus indicating that their information content is very specific. If confirmed with the finalized data, the content of the slow subsequences may finally help resolving the puzzle of slow and fast ribosomal velocity. Furthermore, the deep learning tools can be used to produce a reliable ribosome profile over the entire transcriptome, even at positions where the signal-to-noise-ratio of the experimental data is too low. This tool may finally completely revolutionize the way ribo-seq data is analyzed and provide experimentalists a guide to search for regulatory elements in translational control.
Translational control of gene expression in cancer
Translational control of gene expression is a modern paradigm that has emerged in recent times. While control of gene expression at the level of transcription modulates the amount of proteins by changing the number of mRNA molecules in the cells, translational control targets the work of the ribosomes in one or more of their various steps: from initiation, to elongation, termination and recycling. Some of these effects ca be so subtle that only a detailed ribo-seq analysis can reveal them in terms of coding regions that are strongly differentially covered by ribosomes. In one project carried on by Giorgia Giacomini, a PhD student in the group of Monica Bianchini at the University of Siena, we are comparing the ribosomal profiles of healthy versus cancerous liver cells taken from ten patients. Under the supervision of Davide Chiarugi, Monica Bianchini and myself, Giorgia Giacomini exploits a technique that Davide Chiarugi and I developed some time ago that makes such a comparison very robust. Preliminary results show that only a few genes are under translational control in a systematic way in the cancerous cells compared to the healthy cell. This study may reveal that some characteristics of the cancerous cells are determined by the way their proteins are synthesized more than by the amount of mRNA present in the cells.