Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Networks”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 595 records · Page 33Linked to original sources
PubMed2003

A model for single and multiple knowledge based networks.

The inherent black-box nature of neural networks is an important drawback with respect to the problem of explanation of neural network responses. Although several articles have tackled the problem of rule extraction from a single neural network, just a few papers have investigated rule extraction from several combined neural networks. In this article we describe how to translate symbolic rules into the Discretized Interpretable Multi-Layer Perceptron (DIMLP) and how to extract rules from one or several combined neural networks. Our approach consists of characterizing discriminant hyperplane frontiers. Unordered rules are extracted in polynomial time with respect to the size of the problem and the size of the network. Moreover, the degree of matching between extracted rules and neural network responses is 100% on training examples. We applied single DIMLP networks to 17 data sets related to medical diagnosis and medical prognosis problems. Results based on 10-fold cross-validation showed that the DIMLP model was on average as accurate as standard multi-layer perceptrons (MLP). Furthermore, DIMLP networks were significantly more accurate than CN2 on eight problems, whereas only on one problem CN2 was better than DIMLP. Finally, a non-Hodgkin lymphoma diagnosis problem based on classification of electrophoresis gels was defined. It turned out that ensembles of DIMLP networks were significantly more accurate than CN2 (96.1% +/- 1.4 versus 82.7% +/- 4.0). Finally, symbolic rules revealed the presence of five important spots for the discrimination of the class of Lymphocyte Leukemia/Chronic Lymphoid Leukemia (Lc/LLc), and the class of Centrocytic Lymphoma (Cc).

Algorithms↗
PubMed1995

Artificial neural networks for screening patients needing emergency cranial computed tomography scans in emergency departments.

RATIONALE AND OBJECTIVES: We evaluated the potential for a neural network to screen candidates for emergency cranial computed tomography (CT) scans in an emergency department setting. METHODS: Data were collected from 1625 patients undergoing emergency cranial CT scanning in two different emergency departments (EDs). Singular value decomposition (SVD) was used to remap input data for network training. Data were randomly divided into six subsets, and one was reserved as a test set to analyze network performance. Five networks were then trained on data from the five remaining sets using fivefold cross-validation. Each trained network was allowed an independent vote on need for CT scanning in each case from the test set. The majority vote was used as the final prediction. A similar analysis was done on data from each individual ED. Results are compared with prior statistical studies of the same data. RESULTS: The network performed well when predicting clinical variable patterns that consistently produced negative CT scans and on patterns that were ambiguous in terms of the CT scan results. It performed poorly, however, on patterns that consistently predicted positive scans. This last finding appears to have resulted from inadequate training material. The two populations from which data were taken were shown to be distinct, but a network trained on the combined data performed as well as the networks from the individual EDs in predicting patients requiring CT scanning. Variables with the greatest contribution to the networks' prediction were consistent with those in prior statistical studies. CONCLUSION: Although preliminary in nature, neural networks show promise as a screening device for selecting patients for emergent cranial CT scanning.

Analysis of Variance↗
PubMed1998

Improved diagnosis of breast implant rupture with sonographic findings and artificial neural networks.

RATIONALE AND OBJECTIVES: The authors evaluated the use of sonographic findings combined with artificial neural networks as an aid to the diagnosis of breast implant rupture. MATERIALS AND METHODS: From a database of 78 breast implants that were evaluated prospectively with sonography and then surgically removed, sonographic findings and surgical results were used to train and test backpropagation and radial basis function artificial neural networks by using the leave-one-out method. Receiver operating characteristic (ROC) curve analysis was used to compare the performance of the different neural networks with that of the radiologists involved. RESULTS: By using the ROC area index as a measure of performance, the artificial neural network (Az = 0.8744) outperformed the radiologists (Az = 0.8057), although not by a statistically significant difference (P = .09). The best-performing network used, in addition to the sonographic findings, the diagnosis of the radiologist as an input. This network (Az = 0.9245) outperformed both the radiologists and the "unaided" networks by a statistically significant margin (P = .02 for radiologists, P = .04 for the unaided network). The network performed remarkably well in those cases in which the radiologists classified the implant as indeterminate, predicting the correct diagnosis in 23 of 25 cases (92%). CONCLUSION: The results suggest that artificial neural networks in tandem with the unaided radiologic diagnosis can improve the accuracy rate in the detection of implant rupture based on sonographic findings. This "team" approach provided the best results.

Breast Implants↗
PubMed2000

Scandinavian test of artificial neural network for classification of myocardial perfusion images.

Artificial neural networks are systems of elementary computing units capable of learning from examples. They have been applied to automated interpretation of myocardial perfusion images and have been shown to perform even better than experienced physicians. It has been shown that physicians interpreting myocardial perfusion images benefit from the advice of such networks. These networks have been developed and validated in the same hospital. However, widespread use of neural networks will only take place if the networks can maintain a high accuracy in other hospitals, i.e. hospitals using different gamma cameras, different acquisition techniques, different study protocols, etc. The purpose of this study was to develop a neural network in one hospital and test it in another. An artificial neural network was trained to detect coronary artery disease using myocardial perfusion scintigrams from 135 patients at a Swedish hospital. Thereafter, this network was tested using scintigrams from 68 patients at a Danish hospital and compared to six criteria based on expert physician analysis and quantitative analysis by the CEqual program. The sensitivity of the network was significantly higher than that of one of the physician criteria (0. 92 versus 0.71) and two of the CEqual-based criteria (0.94 versus 0. 63 and 0.96 versus 0.65) compared at equal specificities. It was concluded that an artificial neural network can maintain high accuracy in a hospital other than the one where it was developed.

Coronary Disease↗
PubMed2001

Networked information and clinical decision making: the experience of Birmingham Heartlands and Solihull National Health Service Trust (Teaching).

OBJECTIVES: The paper presents the findings of the evaluation of a pilot project to introduce networked information resources into clinical settings in a large NHS Trust and describes the subsequent developments of networked information within the Trust. DESIGN: The main purpose of the evaluation was to ascertain whether access to electronic journals and other resources via a networked system offered real benefits to clinical effectiveness. SETTING: Birmingham Heartlands and Solihull NHS Trust (Teaching). SUBJECTS: Medical and administrative staff at Birmingham and Solihull NHS Trust. RESULTS: The main conclusions of the evaluation were that: (1) appropriate location of terminals close to clinical areas is vital to ensure that best use is made of networked information resources; (2) rapid access to networked information services saves staff time and allows educational opportunities to be realized; (3) networked information resources enhance rather than replace existing information sources; (4) training and support are essential to maximizing the benefits of networked information services, and (5) such a network can support clinical decision making. CONCLUSIONS: With the development of clinical governance, the clinical network has assumed even greater importance within the Trust. Timely and easy access to clinical and educational information is crucial to the practice of evidence-based medicine which underpins high quality clinical care. The evaluation led to a number of recommendations which have since been used to develop the clinical network at Birmingham Heartlands and Solihull NHS Trust.

Adult↗
PubMed2006

An effective structure learning method for constructing gene networks.

MOTIVATION: Bayesian network methods have shown promise in gene regulatory network reconstruction because of their capability of capturing causal relationships between genes and handling data with noises found in biological experiments. The problem of learning network structures, however, is NP hard. Consequently, heuristic methods such as hill climbing are used for structure learning. For networks of a moderate size, hill climbing methods are not computationally efficient. Furthermore, relatively low accuracy of the learned structures may be observed. The purpose of this article is to present a novel structure learning method for gene network discovery. RESULTS: In this paper, we present a novel structure learning method to reconstruct the underlying gene networks from the observational gene expression data. Unlike hill climbing approaches, the proposed method first constructs an undirected network based on mutual information between two nodes and then splits the structure into substructures. The directional orientations for the edges that connect two nodes are then obtained by optimizing a scoring function for each substructure. Our method is evaluated using two benchmark network datasets with known structures. The results show that the proposed method can identify networks that are close to the optimal structures. It outperforms hill climbing methods in terms of both computation time and predicted structure accuracy. We also apply the method to gene expression data measured during the yeast cycle and show the effectiveness of the proposed method for network reconstruction.

Algorithms↗
PubMed1995

Injury severity and probability of survival assessment in trauma patients using a predictive hierarchical network model derived from ICD-9 codes.

UNLABELLED: Accurate assessment of injury severity is critical for decision making related to the prevention, triage, and treatment of injured patients. Presently, the standard method of controlling for variations of injury severity between groups has been based upon the Injury Severity Score (ISS) and the Trauma Score and the Trauma and Injury Severity Score (TRISS) methodology. The purpose of this study was to attempt to build upon previous work using International Classification of Diseases, ninth revision (ICD-9) coded diagnosis, and procedure information available from standard hospital discharge abstracts (UB-82 Billing format) to create a hierarchical network to provide a tool for predicting injury severity and probability of survival. METHODS: Data were obtained for this analysis from the North Carolina Medical Database. Data were available on all trauma patients admitted to hospitals in North Carolina from January 1, 1988 until June 30, 1992. The dependent variable of interest was the patient's survival after injury, coded as live or die. The independent variables used in the study included the ISS derived using the technique described by MacKenzie Abbreviated Injury Score (AIS) and body system maximum AIS scores, mortality risk ratios derived from the ICD-9-DM primary, secondary, and tertiary diagnoses, primary and secondary procedures as described in previous work, age and gender. Network generation used a commercial software package, AIM (Abtech Corp., Charlottesville, Va.), which is a numeric modeling tool that automatically "learns" knowledge from a data base of examples. RESULTS: In the test data set an ISS and a prediction of survival based upon the derived network were calculated for each and every patient. The relative predictive power of these two scores were compared by calculating the overall accuracy, sensitivity, and specificity and the false positive and false negative rates. The receiver operator characteristic curves demonstrate that the network is a more effective tool in predicting the outcome of trauma patients. All the measures of predictive power show that the network was the better predictor of outcome than the ISS. CONCLUSIONS: Given the recognized limitations of the ISS, the widespread availability of the ICD-9 coded diagnoses and procedures, and the availability of many state and regional data bases that have no ISS or Trauma Score, the purpose of this study was to assess the ability of a network derived from limited but widely available hospital discharge data to predict the outcome of injured patients. The study confirms previous work showing that the ICD-9 codes were strongly associated with outcome. The study demonstrated that the network created from these data was a better predictor of outcome than the derived ISS. When the results of the network were compared with other published series, the network, created without access to physiologic information, was almost as accurate, sensitive, and specific as reported values for TRISS and A Severity Characterization of Trauma (ASCOT). Because the present study is the first of its type, further investigations are needed to validate these findings. If other studies corroborate this study, a network model based upon ICD-9 codes could become the principal method for grading injury severity. This would provide superior predictive power of injury severity with important cost savings and universal application.

Adult↗
PubMed1998

Experience with PACS in an ATM/Ethernet switched network environment.

Legacy local area network (LAN) technologies based on shared media concepts are not adequate for the growth of a large-scale picture archiving and communication system (PACS) in a client-server architecture. First, an asymmetric network load, due to the requests of a large number of PACS clients for only a few main servers, should be compensated by communication links to the servers with a higher bandwidth compared to the clients. Secondly, as the number of PACS nodes increases, the network throughout should not measurably cut production. These requirements can easily be fulfilled using switching technologies. Here asynchronous transfer mode (ATM) is clearly one of the hottest topics in networking because the ATM architecture provides integrated support for a variety of communication services, and it supports virtual networking. On the other hand, most of the imaging modalities are not yet ready for integration into a native ATM network. For a lot of nodes already joining an Ethernet, a cost-effective and pragmatic way to benefit from the switching concept would be a combined ATM/Ethernet switching environment. This incorporates an incremental migration strategy with the immediate benefits of high-speed, high-capacity ATM (for servers and high-sophisticated display workstations), while preserving elements of the existing network technologies. In addition, Ethernet switching instead of shared media Ethernet improves the performance considerably. The LAN emulation (LANE) specification by the ATM forum defines mechanisms that allow ATM networks to coexist with legacy systems using any data networking protocol. This paper points out the suitability of this network architecture in accordance with an appropriate system design.

Local Area Networks↗
PubMed2004

A general projection neural network for solving monotone variational inequalities and related optimization problems.

Recently, a projection neural network for solving monotone variational inequalities and constrained optimization problems was developed. In this paper, we propose a general projection neural network for solving a wider class of variational inequalities and related optimization problems. In addition to its simple structure and low complexity, the proposed neural network includes existing neural networks for optimization, such as the projection neural network, the primal-dual neural network, and the dual neural network, as special cases. Under various mild conditions, the proposed general projection neural network is shown to be globally convergent, globally asymptotically stable, and globally exponentially stable. Furthermore, several improved stability criteria on two special cases of the general projection neural network are obtained under weaker conditions. Simulation results demonstrate the effectiveness and characteristics of the proposed neural network.

Neural Networks, Computer↗
PubMed2005

A primal-dual neural network for online resolving constrained kinematic redundancy in robot motion control.

This paper proposes a primal-dual neural network with a one-layer structure for online resolution of constrained kinematic redundancy in robot motion control. Unlike the Lagrangian network, the proposed neural network can handle physical constraints, such as joint limits and joint velocity limits. Compared with the existing primal-dual neural network, the proposed neural network has a low complexity for implementation. Compared with the existing dual neural network, the proposed neural network has no computation of matrix inversion. More importantly, the proposed neural network is theoretically proved to have not only a finite time convergence, but also an exponential convergence rate without any additional assumption. Simulation results show that the proposed neural network has a faster convergence rate than the dual neural network in effectively tracking for the motion control of kinematically redundant manipulators.

Algorithms↗
PubMed1993

Effects on response time of factors selectively influencing processes in acyclic task networks with OR gates.

The mental processes involved in performing some tasks can be represented as directed arcs in an acyclic network. A path directed from the head of one arc to the tail of another indicates that the process represented by the first arc must be executed prior to the process represented by the second arc. If there is no directed path from one arc to another, the corresponding processes can be executed concurrently. Information about the arrangement of processes in an acyclic network can be found from the effects on response times of factors selectively influencing the processes. The methodology was developed earlier for critical path networks, in which a process begins execution when all its immediate predecessors have finished. This paper considers shortest path networks, in which a process begins execution as soon as any immediate predecessor is finished. Results analogous to those for critical path networks are reported. New results are presented enabling investigators to distinguish sequential and concurrent processes in both critical path and shortest path networks. This information is sufficient to construct an acyclic network representing the processes. Further, by examining the effects of selectively influencing processes, one can determine whether a task network is a critical path network or a shortest path network.

Animals↗
PubMedDate not supplied

An artificial neural network for estimating scatter exposures in portable chest radiography.

An adaptive linear element (Adaline) was developed to estimate the two-dimensional scatter exposure distribution in digital portable chest radiographs (DPCXR). DPCXRs and quantitative scatter exposure measurements at 64 locations throughout the chest were acquired for ten radiographically normal patients. The Adaline is an artificial neural network which has only a single node and linear thresholding. The Adaline was trained using DPCXR-scatter measurement pairs from five patients. The spatially invariant network would take a portion of the image as its input and estimate the scatter content as output. The trained network was applied to the other five images, and errors were evaluated between estimated and measured scatter values. Performance was compared against a convolution scatter estimation algorithm. The network was evaluated as a function of network size, initial values, and duration of training. Network performance was evaluated qualitatively by the correlation of network weights to physical models, and quantitatively by training and evaluation errors. Using DPCXRs as input, the network learned to describe known scatter exposures accurately (7% error) and estimate scatter in new images (< 8% error) slightly better than convolution methods. Regardless of size and initial shape, all networks adapted into radial exponentials with magnitude of 0.75, perhaps implying an ideal point spread function and average scatter fraction, respectively. To implement scatter compensation, the two-dimensional scatter distribution estimated by the neural network is subtracted from the original DPCXR.

Algorithms↗
PubMed1994

Computerized detection of clustered microcalcifications in digital mammograms using a shift-invariant artificial neural network.

A computer-aided diagnosis (CAD) scheme has been developed in our laboratory for the detection of clustered microcalcifications in digital mammograms. In this study, we apply a shift-invariant neural network to eliminate false-positive detections reported by the CAD scheme. The shift-invariant neural network is a multilayer back-propagation neural network with local, shift-invariant interconnections. The advantage of the shift-invariant neural network is that the result of the network is not dependent on the locations of the clustered microcalcifications in the input layer. The neural network is trained to detect each individual microcalcification in a given region of interest (ROI) reported by the CAD scheme. A ROI is classified as a positive ROI if the total number of microcalcifications detected in the ROI is greater than a certain number. The performance of the shift-invariant neural network was evaluated by means of a jackknife (or holdout) method and ROC analysis using a database of 168 ROIs, as reported by the CAD scheme when applied to 34 mammograms. The analysis yielded an average area under the ROC curve (Az) of 0.91. Approximately 55% of false-positive ROIs were eliminated without any loss of the true-positive ROIs. The result is considerably better than that obtained in our previous study using a conventional three-layer, feed-forward neural network. The effect of the network structure on the performance of the shift-invariant neural network is also studied.

Biophysical Phenomena↗
PubMed1994

Relationship between a fuzzy logic and a steepest descent approach to optimize a feedforward artificial neural network configuration.

The neural network designer must take into consideration many factors when selecting an appropriate network configuration. The performance of a given network configuration is influenced by many different factors such as: accuracy, training time, sensitivity, and the number of neurons used in the implementation. Using a cost function based on the four criteria mentioned previously, the various network paradigms can be evaluated relative to one another. If the mathematical models of the evaluation criteria as functions of the network configuration are known, then traditional techniques (such as the steepest descent method) could be used to determine the optimal network configuration. The difficulty in selecting an appropriate network configuration is due to the difficulty involved in determining the mathematical models of the evaluation criteria. This difficulty can be avoided by using fuzzy logic techniques to perform the network optimization as opposed to the traditional techniques. Fuzzy logic avoids the need of a detailed mathematical description of the relationship between the network performance and the network configuration, by using heuristic reasoning and linguistic variables. A comparison will be made between the fuzzy logic approach and the steepest descent method for the optimization of the cost function. The fuzzy optimization procedure could be applied to other systems where there is a priori information about their characteristics.

Electronic Data Processing↗
PubMed1999

A new neural network architecture with associative memory, pruning and order-sensitive learning.

A new paradigm of neural network architecture is proposed that works as associative memory along with capabilities of pruning and order-sensitive learning. The network has a composite structure wherein each node of the network is a Hopfield network by itself. The Hopfield network employs an order-sensitive learning technique and converges to user-specified stable states without having any spurious states. This is based on geometrical structure of the network and of the energy function. The network is so designed that it allows pruning in binary order as it progressively carries out associative memory retrieval. The capacity of the network is 2n, where n is the number of basic nodes in the network. The capabilities of the network are demonstrated by experimenting on three different application areas, namely a Library Database, a Protein Structure Database and Natural Language Understanding.

Artificial Intelligence↗
PubMed1996

Neural network based on the input organization of an identified neuron signaling impending collision.

1. We describe a four-layered neural network (Fig. 1), based on the input organization of a collision signaling neuron in the visual system of the locust, the lobula giant movement detector (LGMD). The 250 photoreceptors ("P" units) in layer 1 are excited by any change in illumination, generated when an image edge passes over them. Layers 2 and 3 incorporate both excitatory and inhibitory interactions, and layer 4 consists of a single output element, equivalent to the locust LGMD. 2. The output element of the neural network, the "LGMD", responds directionally when challenged with approaching versus receding objects, preferring approaching objects (Figs. 2-4). The time course and shape of the "LGMD" response matches that of the LGMD (Fig. 4). Directionality is maintained with objects of various sizes and approach velocities. The network is tuned to direct approach (Fig. 5). The "LGMD" shows no directional selectivity for translatory motion at a constant velocity across the "eye", but its response increases with edge velocity (Figs. 6 and 9). 3. The critical image cues for a selective response to object approach by the "LGMD" are edges that change in extent or in velocity as they move (Fig. 7). Lateral inhibition is crucial to the selectivity of the "LGMD" and the selective response is abolished or else much reduced if lateral inhibition is taken out of the network (Fig. 7). We conclude that lateral inhibition in the neuronal network for the locust LGMD also underlies the experimentally observed critical image cues for its directional response. 4. Lateral inhibition shapes the velocity tuning of the network for objects moving in the X and Y directions without approaching the eye (see Fig. 1). As an edge moves over the eye at a constant velocity, a race occurs between the excitation that is caused by edge movement and which passes down the network and the inhibition that passes laterally. Excitation must win this race for units in layer 3 to reach threshold (Fig. 8). The faster the edge moves over the eye the more units in layer 3 reach threshold and pass excitation on to the "LGMD" (Fig. 9). 5. Lateral inhibition shapes the tuning of the network for objects moving in the Z direction, toward or away from the eye (see Fig. 1). As an object approaches the eye there is a buildup of excitation in the "LGMD" throughout the movement whereas the response to object recession is often brief, particularly for high velocities. During object motion, a critical race occurs between excitation passing down the network and inhibition directed laterally, excitation must win this race for the rapid buildup in excitation in the "LGMD" as seen in the final stages of object approach (Figs. 10-12). The buildup is eliminated if, during object approach, excitation cannot win this race (as happens when the spread of inhibition laterally takes < 1 ms Fig. 13, D and E). Taking all lateral inhibition away increases the "LGMD" response to object approach, but overall directional selectivity is reduced as there is also a lot of residual network excitation following object recession (Fig. 13B). 6. Directional selectivity for rapidly approaching objects is further enhanced at the level of the "LGMD" by the timing of a feed-forward, inhibitory loop onto the "LGMD", activated when a large number of receptor units are excited in a short time. The inhibitory loop is activated at the end of object approach, truncating the excitatory "LGMD" response after approach has ceased, but at the initiation of object recession (*Fig. 2, 3, and 13). Eliminating the feed-forward, inhibitory loop prolongs the "LGMD" response to both receding and approaching objects (Fig. 13F).

Animals↗
PubMed1999

Discontinuities in recurrent neural networks.

This article studies the computational power of various discontinuous real computational models that are based on the classical analog recurrent neural network (ARNN). This ARNN consists of finite number of neurons; each neuron computes a polynomial net function and a sigmoid-like continuous activation function. We introduce arithmetic networks as ARNN augmented with a few simple discontinuous (e.g., threshold or zero test) neurons. We argue that even with weights restricted to polynomial time computable reals, arithmetic networks are able to compute arbitrarily complex recursive functions. We identify many types of neural networks that are at least as powerful as arithmetic nets, some of which are not in fact discontinuous, but they boost other arithmetic operations in the net function (e.g., neurons that can use divisions and polynomial net functions inside sigmoid-like continuous activation functions). These arithmetic networks are equivalent to the Blum-Shub-Smale model, when the latter is restricted to a bounded number of registers. With respect to implementation on digital computers, we show that arithmetic networks with rational weights can be simulated with exponential precision, but even with polynomial-time computable real weights, arithmetic networks are not subject to any fixed precision bounds. This is in contrast with the ARNN that are known to demand precision that is linear in the computation time. When nontrivial periodic functions (e.g., fractional part, sine, tangent) are added to arithmetic networks, the resulting networks are computationally equivalent to a massively parallel machine. Thus, these highly discontinuous networks can solve the presumably intractable class of PSPACE-complete problems in polynomial time.

Algorithms↗
PubMed2005

Improved prediction of critical residues for protein function based on network and phylogenetic analyses.

BACKGROUND: Phylogenetic approaches are commonly used to predict which amino acid residues are critical to the function of a given protein. However, such approaches display inherent limitations, such as the requirement for identification of multiple homologues of the protein under consideration. Therefore, complementary or alternative approaches for the prediction of critical residues would be desirable. Network analyses have been used in the modelling of many complex biological systems, but only very recently have they been used to predict critical residues from a protein's three-dimensional structure. Here we compare a couple of phylogenetic approaches to several different network-based methods for the prediction of critical residues, and show that a combination of one phylogenetic method and one network-based method is superior to other methods previously employed. RESULTS: We associate a network with each member of a set of proteins for which the three-dimensional structure is known and the critical residues have been previously determined experimentally. We show that several network-based centrality measurements (connectivity, 2-connectivity, closeness centrality, betweenness and cluster coefficient) accurately detect residues critical for the protein's function. Phylogenetic approaches render predictions as reliable as the network-based measurements, although, interestingly, the two general approaches tend to predict different sets of critical residues. Hence we propose a hybrid method that is composed of one network-based calculation--the closeness centrality--and one phylogenetic approach--the Conseq server. This hybrid approach predicts critical residues more accurately than the other methods tested here. CONCLUSION: We show that network analysis can be used to improve the prediction of amino acids critical for protein function, when utilized in combination with phylogenetic approaches. It is proposed that such improvement is due to the complementary nature of these approaches: network-based methods tend to predict as critical those residues that are highly connected and internal (i.e., non-surface), although some surface residues are indeed identified as critical by network analyses; whereas residues chosen by phylogenetic approaches display a lower overall probability of being surface inaccessible.

Computational Biology↗