Database approaches are widely used in structural bioinformatics, since ab initio techniques are often computationally prohibitive, and the structure of biological macromolecules are typically derived from a limited set of motifs. There are several issues and challenges that arise when developing methods to enable efficient database retrieval. For example, how can complex data be represented efficiently, and what should be the size and composition of the database? In this work, we discuss some of these challenges, based on a crystallographic protein model-building program called TEXTAL. In particular, we discuss how structural information on amino acids is represented (as numeric features), how difficult it is to recognize amino acids (based on 3D electron density patterns), and what types of examples (and how many of them) need to be stored in the database. These insights are potentially useful in many other related applications, such as structure-based drug design, protein-protein interaction, discriminating nucleic acids and proteins in hybrid complexes, etc.

Data preprocessing is critical for machine learning, data mining, and pattern recognition. In particular, selecting relevant and non-redundant features in high-dimensional data is important to efficiently construct models that accurately describe the data. In this work, I present SLIDER, an algorithm that weights features to reflect relevance in determining similarity between instances. Accurate weighting of features improves the similarity measure, which is useful in learning algorithms like nearest neighbor and case-based reasoning. SLIDER performs a greedy search for optimum weights in an exponentially large space of weight vectors. Exhaustive search being intractable, the algorithm reduces the search space by focusing on pivotal weights at which representative instances are equidistant to truly similar and different instances in Euclidean space. SLIDER then evaluates those weights heuristically, based on effectiveness in properly ranking pre-determined matches of a set of cases, relative to mismatches. I analytically show that by choosing feature weights that minimize the mean rank of matches relative to mismatches, the separation between the distributions of Euclidean distances for matches and mismatches is increased. This leads to a better distance metric, and consequently increases the probability of retrieving true matches from a database. I also discuss how SLIDER is used to improve the efficiency and effectiveness of case retrieval in a case-based reasoning system that automatically interprets electron density maps to determine the three-dimensional structures of proteins. Electron density patterns for regions in a protein are represented by numerical features, which are used in a distance metric to efficiently retrieve matching patterns by searching a large database. These pre-selected cases are then evaluated by more expensive methods to identify truly good matches—this strategy speeds up the retrieval of matching density regions, thereby enabling fast and accurate protein model-building. This two-phase case retrieval approach is potentially useful in many case-based reasoning systems, especially those with computationally expensive case matching and large case libraries.

TEXTAL is a computer program that automatically-interprets electron density maps to determine the atomic structures of proteins through X-ray crystallography. Electron density maps are traditionally interpreted by visually fitting atoms into density Patterns. This manual process can be time-consuming and error prone, even for expert crystallographers, Noise in the data and limited resolution make map interpretation challenging. To automate the process, TEXTAL employs a variety of AI and Pattern-recognition techniques that emulate the decision-making processes of domain experts. In this article, we discuss the various ways AI technology is used in TEXTAL, including neural networks, case-based reasoning, nearest neighbor learning and linear discriminant analysis. The AI and pattern-recognition approaches have proven to be effective for building protein models even with medium resolution data. TEXTAL is a successfully deployed application; it is being used in more than 100 crystallography labs from 20 countries.

UNLABELLED X-ray crystallography is the most widely used method to determine the 3D structure of protein molecules. One of the most difficult steps in protein crystallography is model-building, which consists of constructing a backbone and then amino acid side chains into an electron density map. Interpretation of electron density maps represents a major bottleneck in protein structure determination pipelines, and thus, automated techniques to interpret maps can greatly improve the throughput. We have developed WebTex, a simple and yet powerful web interface to TEXTAL, a program that automates this process of fitting atoms into electron density maps. TEXTAL can also be downloaded for local installation. AVAILABILITY Web interface, downloadable binaries and documentation at http://textal.tamu.edu

A new software system called PHENIX (Python-based Hierarchical ENvironment for Integrated Xtallography) has been developed for the automation of crystallographic structure solution. This provides algorithms to go from reduced intensity data to a refined molecular model, and facilitates structure solution for both the novice and expert crystallographer. Here, we review the major features of PHENIX, including the different user interfaces, and briefly describe the recent advances in infrastructure and algorithms.

TEXTAL is a successfully deployed system for automated model-building in protein X-ray crystallography. It represents a novel solution to an important, complex real-world, problem using various AI and pattern recognition algorithms. TEXTAL takes a model-building approach based on real-space density pattern recognition, similar to how a human crystallographer would work. TEXTAL first tries to predict the coordinates of the alpha-carbon (C/spl alpha/) atoms in the protein's connected backbone using a neural network. It then analyzes the density patterns around each C/spl alpha/ atom and searches a database of previously solved structures for regions with similar patterns. TEXTAL determines the best match, retrieves the coordinates for that region, and fits them to the unknown density. TEXTAL concatenates these local models into a global model and subjects them to various subsequent refinements to produce a complete protein model automatically.

High-throughput computational methods in X-ray protein crystallography are indispensable to meet the goals of structural genomics. In particular, automated interpretation of electron density maps, especially those at mediocre resolution, can significantly speed up the protein structure determination process. TEXTALTM is a software application that uses pattern recognition, case-based reasoning and nearest neighbor learning to produce reasonably refined molecular models, even with average quality data. In this work, we discuss a key issue to enable fast and accurate interpretation of typically noisy electron density data: what features should be used to characterize the density patterns, and how relevant are they? We discuss the challenges of constructing features in this domain, and describe SLIDER, an algorithm to determine the weights of these features. SLIDER searches a space of weights using ranking of matching patterns (relative to mismatching ones) as its evaluation function. Exhaustive search being intractable, SLIDER adopts a greedy approach that judiciously restricts the search space only to weight values that cause the ranking of good matches to change. We show that SLIDER contributes significantly in finding the similarity between density patterns, and discuss the sensitivity of feature relevance to the underlying similarity metric.

This paper reports on TEXTAL™, a deployed application that uses a variety of AI techniques to automate the process of determining the 3D structure of proteins by x-ray crystallography. The TEXTAL™ project was initiated in 1998, and the application is currently deployed in three ways: (1) a web-based interface called WebTex, operational since June 2002; (2) as the automated model-building component of an integrated crystallography software called PHENIX, first released in July 2003; (3) binary distributions, available since September 2004. TEXTAL™ and its sub-components are currently being used by crystallographers around the world, both in the industry and in academia. TEXTAL™ saves up to weeks of effort typically required to determine the structure of one protein; the system has proven to be particularly helpful when the quality of the data is poor, which is very often the case. Automated protein modeling systems like TEXTAL™ are critical to the structural genomics initiative, a worldwide effort to determine the 3D structure of all proteins in a high-throughput mode, thereby keeping up with the rapid growth of genomic sequence databases.

Feature selection and weighting are central problems in pattern recognition and instance-based learning. In this work, we discuss the challenges of constructing and weighting features to recognize 3D patterns of electron density to determine protein structures. We present SLIDER, a feature-weighting algorithm that adjusts weights iteratively such that patterns that match query instances are better ranked than mismatching ones. Moreover, SLIDER makes judicious choices of weight values to be considered in each iteration, by examining specific weights at which matching and mismatching patterns switch as nearest neighbors to query instances. This approach reduces the space of weight vectors to be searched. We make the following two main observations: (1) SLIDER efficiently generates weights that contribute significantly in the retrieval of matching electron density patterns; (2) the optimum weight vector is sensitive to the distance metric i.e. feature relevance can be, to a certain extent, sensitive to the underlying metric used to compare patterns.

A new software system called PHENIX (Python-based Hierarchical ENvironment for Integrated Xtallography) is being developed for the automation of crystallographic structure solution. This will provide the necessary algorithms to proceed from reduced intensity data to a refined molecular model, and facilitate structure solution for both the novice and expert crystallographer. Here we review the features of PHENIX and briefly describe the recent advances in infrastructure and algorithms.

Similarity between cases in pattern recognition is typically measured by computing distances between feature vectors. This paper evaluates the effectiveness of various measures of similarity in retrieving good matches in TEXTAL(TM), a system that uses nearest neighbor learning to retrieve matching 3D patterns of electron density to incrementally determine the structure of proteins by X-ray crystallography. We investigate various geometric measures of similarity, including Euclidean, Manhattan (city-block, or L-1), the generalized Minkowsky metric (L-m) and the Cosine measure. We also experiment with a probabilistic distance metric - a likelihood measure based on the Bayesian classifier. Our experiments in the protein crystallography domain show that the probabilistic measure of similarity outperforms geometric ones significantly. We present a general framework for efficient pattern retrieval from a large database using feature-based matching, and argue that probabilistic and statistical measures of similarity are more robust in noisy, high-dimensional feature spaces representing visual patterns.

Inefficient case retrieval is a major problem in many case-based reasoning systems, especially when case matching is expensive and the case-base is large. In this paper, we present a two-phase approach where an inexpensive feature-based method is used to jind a set of potential matches and a more expensive and accurate case matching method is used to make the jinal selection. This approach has been successfully employed in TEXTALTM, a system that retrieves previously solved 3D patterns of electron density from a database to determine the structure of proteins. Electron density patterns are characterized by numeric features and an appropriate distance measure is used to efficiently jilter good matches through an exhaustive search of the database. These matches are then examined using a computationally expensive density correlation procedure based on jinding an optimal superposition between 3D patterns. We provide an empirical and theoretical analysis of some of the keys issues related to this method. In particular, we dejine a model for estimating how approximate various featurebased similarity measures are (relative to an objective matching metric), and determine its relation to the number of cases that should be jiltered from a given database to make the approach effective.

X-ray crystallography is the most widely used method for determining the three-dimensional structures of proteins and other macromolecules. One of the most difficult steps in crystallography is interpreting the 3D image of the electron density cloud surrounding the protein. This is often done manually by crystallographers and is very time-consuming and error-prone. The difficulties stem from the fact that the domain knowledge required for interpreting electron density data is uncertain. Thus crystallographers often have to resort to intuitions and heuristics for decision-making. The problem is compounded by the fact that in most cases, data available is noisy and blurred. TEXTAL ™ is a system designed to automate this challenging process of inferring the atomic structure of proteins from electron density data. It uses a variety of AI and pattern recognition techniques to try to capture and mimic the intuitive decision-making processes of experts in solving protein structures. The system has been quite successful in determining various protein structures, even with average quality data. The initial structure built by TEXTAL ™ can be used for subsequent manual refinement by a crystallographer, and combined with post-processing routines to generate a more complete model.

Saving and reusing previously constructed plans is largely regarded as a promising approach to deal with the intractability of domain-independent planning (Hammond 1989; Kambhampati and Hendler 1992). But it has been shown that syntactically matching a new problem with a candidate case is NP-hard, and modifying a plan to suit a new problem can be strictly more difficult than generating a plan from scratch (Nebel and Koehler 1995). We present a case-based planning system that does not involve any plan modification and performs case matching very efficiently.

The Internet’s explosive growth is making it harder and harder to harness its potential. However, the field of knowledge representation, particularly the subfield of ontologies, can provide techniques for improving the ability of agents to work with Internet information. SHOE (Simple HTML Ontology Extensions) is a semantic markup language designed specifically for the Internet. It includes features that allow knowledge representation in distributed enviroments, and since the Internet is dynamic, allows ontologies to evolve in a controlled way.