Debora Slanzi

Do Natural Proteins Differ from Random Sequences Polypeptides? Natural vs. Random Proteins Classification Using an Evolutionary Neural Network

Are extant proteins the exquisite result of natural selection or are they random sequences slightly edited by evolution? This question has puzzled biochemists for long time and several groups have addressed this issue comparing natural protein sequences to completely random ones coming to contradicting conclusions. Previous works in literature focused on the analysis of primary structure in an attempt to identify possible signature of evolutionary editing. Conversely, in this work we compare a set of 762 natural proteins with an average length of 70 amino acids and an equal number of completely random ones of comparable length on the basis of their structural features. We use an ad hoc Evolutionary Neural Network Algorithm (ENNA) in order to assess whether and to what extent natural proteins are edited from random polypeptides employing 11 different structure-related variables (i.e. net charge, volume, surface area, coil, alpha helix, beta sheet, percentage of coil, percentage of alpha helix, percentage of beta sheet, percentage of secondary structure and surface hydrophobicity). The ENNA algorithm is capable to correctly distinguish natural proteins from random ones with an accuracy of 94.36%. Furthermore, we study the structural features of 32 random polypeptides misclassified as natural ones to unveil any structural similarity to natural proteins. Results show that random proteins misclassified by the ENNA algorithm exhibit a significant fold similarity to portions or subdomains of extant proteins at atomic resolution. Altogether, our results suggest that natural proteins are significantly edited from random polypeptides and evolutionary editing can be readily detected analyzing structural features. Furthermore, we also show that the ENNA, employing simple structural descriptors, can predict whether a protein chain is natural or random.

Designing Lead Optimisation of MMP-12 Inhibitors

The design of new molecules with desired properties is in general a very difficult problem, involving heavy experimentation with high investment of resources and possible negative impact on the environment. The standard approach consists of iteration among formulation, synthesis, and testing cycles, which is a very long and laborious process. In this paper we address the so-called lead optimisation process by developing a new strategy to design experiments and modelling data, namely, the evolutionary model-based design for optimisation (EDO). This approach is developed on a very small set of experimental points, which change in relation to the response of the experimentation according to the principle of evolution and insights gained through statistical models. This new procedure is validated on a data set provided as test environment by Pickett et al. (2011), and the results are analysed and compared to the genetic algorithm optimisation (GAO) as a benchmark. The very good performance of the EDO approach is shown in its capacity to uncover the optimum value using a very limited set of experimental points, avoiding unnecessary experimentation.

Combining Probabilistic Dependency Models and Particle Swarm Optimization for Parameter Inference in Stochastic Biological Systems

In this work we present an efficient method to tackle the problem of parameter inference for stochastic biological models. We develop a variant of the Particle Swarm Optimization algorithm by including Probabilistic Dependency statistical models to detect the parameter dependencies. This results in a more efficient parameter inference of the biological model.We test the Probabilistic Dependency- PSO on a well-known benchmark problem: the thermal isomerization of α-pinene

Detection of volatile metabolites of moulds isolated from a contaminated library

The principal fungal species isolated from a contaminated library environment were tested for their microbial volatile organic compound (MVOC) production ability. Aspergillus creber, A. penicillioides, Cladosporium cladosporioides, Eurotium chevalieri, E. halophilicum, Penicillium brevicompactum and P. chrysogenum were cultivated on suitable culture media inside sample bottles specifically designed and created for direct MVOC injection to a GC-MS instrument. The fungal emissions were monitored over several weeks to detect changes with the aging of the colonies, monitored also by respirometric tests. A total of 55 different MVOCs were detected and isopropyl alcohol, 3-methyl-1-butanol and 2-butanone were the principal compounds in common between the selected fungal species. Moreover, 2,4-dimethylheptane, 1,4-pentadiene, styrene, ethanol, 2-methyl-1-butanol, acetone, furan and 2-methylfuran were the most detected compounds. For the first time, the MVOC production for particular fungal species was detected. The species A. creber, which belongs to the recently revised group Aspergillus section Versicolores, was characterized by the production of ethanol, furan and 1,4-pentadiene. For the xerophilic fungus E. halophilicum, specific production of acetone, 2-butanone and 1,4-pentadiene was detected, supported also by respirometric data. The results demonstrated the potential use of this method for the detection of fungal contamination phenomena inside Cultural Heritages preservation environments.

On using Bayesian networks for complexity reduction in decision trees

In this paper we use the Bayesian network as a tool of explorative analysis: its theory guarantees that, given the structure and some assumptions, the Markov blanket of a variable is the minimal conditioning set through which the variable is independent from all the others. We use the Markov blanket of a target variable to extract the relevant features for constructing a decision tree (DT). Our proposal reduces the complexity of the DT so it has a simpler visualization and it can be more easily interpretable. On the other hand, it maintains a good classification performance.

Model-Based Lead Molecule Design

“Lead molecule” is a chemical compound deemed as a good candidate for drug discovery. Designing a lead molecule for optimization involves a complex phase in which researchers look for compounds that satisfy pharmaceutical properties and can then be investigated for drug development and clinical trials. Finding the optimal lead molecule is a hard problem that commonly requires searching in high dimensional and large experimental spaces. In this paper we propose to discover the optimal lead molecule by developing an evolutionary model-based approach where different classes of statistical models can achieve relevant information. The analysis is conducted comparing two different chemical representations of molecules: the amino-boronic acid representation and the chemical fragment representation. To deal with the high dimensionality of the fragment representation we adopt the Formal Concept Analysis and we then derive the evolutionary path on a reduced number of fragments. This approach has been tested on a particular data set of 2500 molecules and the achieved results show the very good performance of this strategy.

Reducing Dimensionality in Molecular Systems: A Bayesian Non-parametric Approach

In this paper we present a methodology that can be used to design experiments of complex systems characterized by a huge number of variables. The strategy combines the evolutionary principles with the information provided by statistical models tailored to the problem under consideration. Here, we are concerned with the process of design molecules, which is a quite challenging problem due to the presence of a high number of variables with a binary structure. Recent works on clustering of binary data and variable selection in the high-dimensional setting allow to develop an approach capable of recovering useful information derived from the incorporation of a grouping structure into the model.

Qualitative Particle Swarm Optimization (Q-PSO) for Energy-Efficient Building Designs

Particle Swarm Optimization (PSO) is a stochastic optimization method, based on the social behavior of bird flocks. The method, known for its high performance in optimization, has been mainly developed for problems involving just quantitative variables. In this paper we propose a new approach called Qualitative Particle Swarm Optimization (Q-PSO) where the variables in the optimization can be both qualitative and quantitative and the updating rule is derived adopting probabilistic measures. We apply this procedure to a complex engineering optimization problem concerning building facade design. More specifically, we address the problem of deriving an energy-efficient building design, i.e. a design that minimizes the energy consumption (and the emission of carbon dioxide) for heating, cooling and lighting. We develop a simulation study to evaluate Q-PSO procedure and we derive comparisons with most conventional approaches. The study shows a very good performance of our approach in achieving the assigned target.

Several Computational Studies About Variable Selection for Probabilistic Bayesian Classifiers

The Bayesian network can be considered as a probabilistic classifier with the ability of giving a clear insight into the structural relationships in the domain under investigation. In this paper we use some methodologies of feature subset selection in order to determine the relevant variables which are then used for constructing the Bayesian network. To test how the selected methods of feature selection affect the classification, we consider several Bayesian classifiers: Naive Bayes, Tree Augmented Naive Bayes and the general Bayesian network, which is used as benchmark for the comparison.

Multi-objective Optimization in High-Dimensional Molecular Systems

The paper proposes a methodological approach to design complex experiments for multi-objective optimization. The strategy is based on evolutionary statistical inference to search for the optimal values in high-dimensional experimental spaces. We developed this approach to study a particular molecular system and discover the best molecules to be proposed as candidate drugs.