Long-term cognitive decline prediction based on multi-modal data using Multimodal3DSiameseNet: transfer learning from Alzheimer's disease to Parkinson's disease.

PURPOSE: Monitoring and predicting the cognitive state of subjects with neurodegenerative disorders is crucial to provide appropriate treatment as soon as possible. In this work, we present a machine learning approach using multimodal data (brain MRI and clinical) from two early medical visits, to predict the longer-term cognitive decline of patients. Using transfer learning, our model can be successfully transferred from one neurodegenerative disease (Alzheimer's) to another (Parkinson's). METHODS: Our model is a Deep Neural Network with siamese sub-modules dedicated to extracting features from each modality. We pre-train it with data from ADNI (Alzheimer's disease), then transfer it on the smaller PPMI dataset (Parkinson's disease). We show that, even when we do not fine-tune the filters learnt from the ADNI MRIs, the transferred model's results are satisfying on PPMI. RESULTS: The first main result is that our model provides satisfying long-term predictions of cognitive decline from any pair of early visits, with no fixed time delay between these visits (provided the potential decline has started at the second visit). The second main result is that the prediction performance on Parkinson's dataset (PPMI) reaches an AUC of 0.81 on PPMI after transfer learning from Alzheimer's dataset (ADNI), without even having to re-train the image filters, versus an AUC of 0.72 for the model trained from scratch on PPMI. CONCLUSIONS: First, our model is effective for predicting long-term cognitive decline from only two visits, even with irregular intervals of time. When dealing with neurodegenerative diseases, where patients often miss some control visits, this is an important finding. Second, our model is able to transfer the knowledge learnt from one neurodegenerative disease (Alzheimer's) to another (Parkinson's), when using the same imaging modalities (brain MRI) and different clinical variables. This makes it usable even for diseases that are rare or under-studied.

Metabolic network reconstruction and their topological analysis.

This chapter focuses on the way to build a metabolic network and how to analyze its structure. The first part of this chapter describes the methods of the network model reconstruction from biochemical data found in specialized databases and/or literature. The second part deals with metabolic pathway analysis as a useful tool for better understanding the complex architecture of intracellular metabolism. The graph analysis and the stoichiometric network analysis are important approaches for understanding the network topology and consequently the function of metabolic networks. Among the methods presented, the Elementary Flux Modes analysis will be more detailed. Finally, we illustrate in this chapter an example of network reconstruction from heterotrophic plant cells metabolism and its topological analysis leading to a huge number of Elementary Flux Modes.

Fast computation of minimal cut sets in metabolic networks with a Berge algorithm that utilizes binary bit pattern trees.

Minimal cut sets are a valuable tool for analyzing metabolic networks and for identifying optimal gene intervention strategies by eliminating unwanted metabolic functions and keeping desired functionality. Minimal cut sets rely on the concept of elementary flux modes, which are sets of indivisible metabolic pathways under steady-state condition. However, the computation of minimal cut sets is nontrivial, as even medium-sized metabolic networks with just 100 reactions easily have several hundred million elementary flux modes. We developed a minimal cut set tool that implements the well-known Berge algorithm and utilizes a novel approach to significantly reduce the program run time by using binary bit pattern trees. By using the introduced tree approach, the size of metabolic models that can be analyzed and optimized by minimal cut sets is pushed to new and considerably higher limits.

Comparison between elementary flux modes analysis and 13C-metabolic fluxes measured in bacterial and plant cells.

BACKGROUND: (13)C metabolic flux analysis is one of the pertinent ways to compare two or more physiological states. From a more theoretical standpoint, the structural properties of metabolic networks can be analysed to explore feasible metabolic behaviours and to define the boundaries of steady state flux distributions. Elementary flux mode analysis is one of the most efficient methods for performing this analysis. In this context, recent approaches have tended to compare experimental flux measurements with topological network analysis. RESULTS: Metabolic networks describing the main pathways of central carbon metabolism were set up for a bacteria species (Corynebacterium glutamicum) and a plant species (Brassica napus) for which experimental flux maps were available. The structural properties of each network were then studied using the concept of elementary flux modes. To do this, coefficients of flux efficiency were calculated for each reaction within the networks by using selected sets of elementary flux modes. Then the relative differences - reflecting the change of substrate i.e. a sugar source for C. glutamicum and a nitrogen source for B. napus - of both flux efficiency and flux measured experimentally were compared. For both organisms, there is a clear relationship between these parameters, thus indicating that the network structure described by the elementary flux modes had captured a significant part of the metabolic activity in both biological systems. In B. napus, the extension of the elementary flux mode analysis to an enlarged metabolic network still resulted in a clear relationship between the change in the coefficients and that of the measured fluxes. Nevertheless, the limitations of the method to fit some particular fluxes are discussed. CONCLUSION: This consistency between EFM analysis and experimental flux measurements, validated on two metabolic systems allows us to conclude that elementary flux mode analysis could be a useful tool to complement (13)C metabolic flux analysis, by allowing the prediction of changes in internal fluxes before carbon labelling experiments.

Formal TCA cycle description based on elementary actions.

Many databases propose their own structure and format to provide data describing biological processes. This heterogeneity contributes to the difficulty of large systematic and automatic functional comparisons. To overcome these problems, we have used the BioPsi formal description scheme which allows multi-level representations of biological process information. Applied to the description of the tricarboxylic acid cycle (TCA), we show that BioPsi allows the formal integration of functional information existing in current databases and make them available for further automated analysis. In addition such a formal TCA cycle process description leads to a more accurate biological process annotation which takes in account the biological context. This enables us to perform an automated comparison of the TCA cycles for seven different species based on processes rather than protein sequences. From current databases, BioPsi is able to unravel information that are already known by the biologists but are not available for automated analysis tools and simulation software, because of the lack of formal process descriptions. This use of the BioPsi description scheme to describe the TCA cycle was a key step of the MitoScop project that aims to describe and simulate mitochondrial metabolism in silico.

A hyperstructure approach to mitochondria.

Several questions in our understanding of mitochondria are unanswered. These include how the ratio of mitochondrial (mt)DNA to mitochondria is maintained, how the accumulation of defective, rapidly replicating mitochondrial DNA is avoided, how the ratio of mitochondria to cells is adjusted to fit cellular needs, and why any proteins are synthesized in mitochondria rather than simply imported. In bacteria, large hyperstructures or assemblies of proteins, mRNA, lipids and ions have been proposed to constitute a level of organization intermediate between macromolecules and whole cells. Here, we suggest how the concept of hyperstructures together with other concepts developed for bacteria such as transcriptional sensing and spontaneous segregation may provide answers to mitochondrial problems. In doing this, we show how the problem of the very existence of mtDNA brings its own solution.