Gisbert Schneider

Development of artificial neural filters for pattern recognition in protein sequences

SummaryFour different artificial neural network architectures have been tested for their suitability to extract and predict sequence features. For optimization of the network weights an evolutionary computing method has been applied. The networks have feedforward architecture and provide adaptive neural filter systems for pattern recognition in primary structures and sequence classification. The recognition and prediction of signal peptidase cleavage sites ofE. coli periplasmic protein precursors serves as an example for filter development. The primary structures are represented by seven physicochemical residue properties. This amino acid description provides the feature space for network optimization. The properties hydrophobicity, hydrophilicity, side-chain volume, and polarity allowed an accurate classification of the data. A three-layer network architecture reached a learning success of 100%; the highest prediction accuracy in an independent test set of sequences was 97%. This network architecture appears to be most suited for the analysis ofE. coli signal peptidase cleavage sites. Further suggestions about the design and future applications of artificial neural networks for protein sequence analysis are made.

Peptide design in machina: development of artificial mitochondrial protein precursor cleavage sites by simulated molecular evolution

Artificial neural networks were used for extraction of characteristic physiochemical features from mitochondrial matrix metalloprotease target sequences. The amino acid properties hydrophobicity and volume were used for sequence encoding. A window of 12 residues was employed, encompassing positions -7 to +5 of precursors with cleavage sites. Two sets of noncleavage site examples were selected for network training which was performed by an evolution strategy. The weight vectors of the optimized networks were visualized and interpreted by Hinton diagrams. A neural filter system consisting of 13 perceptron-type networks accurately classified the data. It served as the fitness function in a simulated molecular evolution procedure for sequence-oriented de novo design of idealized cleavage sites. A detailed description of the strategy is given. Several putative high-quality cleavage sites were obtained revealing the critical nature of the residues in the positions -2 and -5. Charged residues seem to have a major influence on cleavage site function.

Development of simple fitness landscapes for peptides by artificial neural filter systems

The applicability of artificial neural filter systems as fitness functions for sequence-oriented peptide design was evaluated. Two example applications were selected: classification of dipeptides according to their hydrophobicity and classification of proteolytic cleavage-sites of protein precursor sequences according to their mean hydrophobicities and mean side-chain volumes. The cleavage-sites covered 12 residues. In the dipeptide experiments the objective was to separate a selected set of molecules from all other possible dipeptide sequences. Perceptrons, feedforward networks with one hidden layer, and a hybrid network were applied. The filters were trained by a (1,λ) evolution strategy. Two types of network units employing either a sigmoidal or a unimodal transfer function were used in the feedforward filters, and their influence on classification was investigated. The two-layer hybrid network employed gaussian activation functions. To analyze classification of the different filter systems, their output was plotted in the two-dimensional sequence space. The diagrams were interpreted as fitness landscapes qualifying the markedness of a characteristic peptide feature which can be used as a guide through sequence space for rational peptide design. It is demonstrated that the applicability of neural filter systems as a heuristic method for sequence optimization depends on both the appropriate network architecture and selection of representative sequence data. The networks with unimodal activation functions and the hybrid networks both led to a number of local optima. However, the hybrid networks produced the best prediction results. In contrast, the filters with sigmoidal activation produced good reclassification results leading to fitness landscapes lacking unreasonable local optima. Similar results were obtained for classification of both dipeptides and cleavage-site sequences.

Evolutionary optimization in multimodal search space

Optimization by a simple evolution strategy based on a mutation and selection scheme without recombination was tested for its efficiency in multimodal search space. A modified Rastrigin function served as an objective function providing fitness landscapes with many local optima. It turned out that the evolutionary algorithm including adaptive stepsize control is wellsuited for optimization. The process is able to efficiently surmount local energy barriers and converge to the global optimum. The relation between the optimization time available and the optimal number of offspring was investigated and a simple rule proposed. Several numbers of offspring are nearly equally suited in a smooth search space, whereas in rough fitness landscapes an optimum is observed. In either case both very large and very small numbers of offspring turned out to be unfavourable for optimization.

Structure optimization of an artificial neural filter detecting membrane‐spanning amino acid sequences

An artificial neural network has been developed for the recognition and prediction of transmembrane regions in the amino acid sequences of human integral membrane proteins. It provides an additional prediction method besides the common hydrophobicity analysis by statistical means. Membrane/nonmembrane transition regions are predicted with 92% accuracy in both training and independent test data. The method used for the development of the neural filter is the algorithm of structure evolution. It subjects both the architecture and parameters of the system to a systematical optimization process and carries out local search in the respective structure and parameter spaces. The training technique of incomplete induction as part of the structure evolution provides for a comparatively general solution of the problem that is described by input-output relations only. Seven physiochemical side-chain properties were used to encode the amino acid sequences. It was found that geometric parameters like side-chain volume, bulkiness, or surface area are of minor importance. The properties polarity, refractivity, and hydrophobicity, however, turned out to support feature extraction. It is concluded that membrane transition regions in proteins are encoded in sequences as a characteristic feature based on the respective side-chain properties. The method of structure evolution is described in detail for this particular application and suggestions for further development of amino acid sequence filters are made.

De novo design of peptides and proteins: machine-generated sequences by the PROSA program

Development of automatic routines for rational protein engineering and design will greatly facilitate many current approaches that are based on rather time-consuming trial-anderror procedures (Thornton, 1992). A new computer-based technique for statistical analysis of sequence features and de novo design of amino acid sequences has been developed: PROSA (PROtein Sequence Analysis) (Schneider and Wrede, 1993). This technique can be applied to the analysis of physico-chemical amino acid sequence properties, generating a multi-dimensional property histogram that will provide the basis for subsequent design of idealized sequences. The first analysis and design experiments with PROSA have already been successful and have led to biologically active sequences (unpublished data). The design is sequence-oriented, i.e. no knowledge of corresponding three-dimensional structures is necessary. A complete description of the PROSA-design algorithm and the original application—the design of leader peptidase cleavage-sites of Escherichia coli periplasmic protein precursors—is described (Schneider and Wrede, 1993). In general, the PROSA procedure consists of four steps:

Zielgerichtetes Design von Aminosäuresequenzen mit Künstlichen Neuronalen Netzen

Es wird eine allgemeine Methode zur Entwicklung von neuronalen Filtersystemen zur Mustererkennung in Proteinsequenzen und fur deren zielgerichtetes Design vorgestellt. Das Verfahren der Pro tein Fi lterinduktion (PROFI) wird am Beispiel der Filterentwicklung zur Erkennung von Signalpeptidase-Schnittstellen veranschaulicht. Mit diesen neuronalen Filtersystemen werden auch Schnittstellen in unabhangigen Testsequenzen mit absoluter Genauigkeit vorhergesagt. Dies stellt eine deutliche Verbesserung gegenuber den bislang verwendeten statistischen Verfahren dar. Der entscheidende Unterschied des PROFI-Systems gegenuber anderen kunstlichen neuronalen Netzen zur Untersuchung von Proteinsequenzen liegt in der biologisch orientierten Datenreprasentation: Nicht Buchstabenfolgen, sondern biophysikalische Aminosaureeigenschaften der Sequenzen werden auf charakteristische Merkmale hin untersucht. Die Merkmalssuche kann man als eine Optimierungsaufgabe betrachten. Als Optimierungsverfahren wurde die Evolutionsstrategie angewandt. Mit diesem neuen Verfahren konnen sich in Zukunft einige Decodierungsprobleme in Proteinsequenzen losen lassen. Die mit PROFI entwickelten neuronalen Filtersysteme werden zur Prozesskontrolle beim rationalen Design von Aminosauresequenzen eingesetzt. Dazu wird ein Protein-Design Zyklus vorgestellt, der die Optimierung von Aminosauresequenzen nach diesem kybernetischen Modell ermoglicht (PROSID: Pro tein S equence I nductive D esign). Trainierte kunstliche neuronale Netze dienen dabei zur Reprasentation eines „idealen“ Proteinmodells, welches in einer sich wiederholenden Mutations-Selektions Prozedur aus einer zuachst zufallig gewahlten Aminosauresequenz durch S imulierte M olekulare E volution (SME) erzeugt werden kann. Dieses neue Syntheseverfahren ist als Prototyp fur DOS-Rechner in Modula2 implementiert.