WO2002065119A9 - A method for the prediction of molecular interaction networks - Google Patents
A method for the prediction of molecular interaction networksInfo
- Publication number
- WO2002065119A9 WO2002065119A9 PCT/US2002/004028 US0204028W WO02065119A9 WO 2002065119 A9 WO2002065119 A9 WO 2002065119A9 US 0204028 W US0204028 W US 0204028W WO 02065119 A9 WO02065119 A9 WO 02065119A9
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- interaction
- proteins
- probability
- network
- probabilities
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B35/00—ICT specially adapted for in silico combinatorial libraries of nucleic acids, proteins or peptides
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B35/00—ICT specially adapted for in silico combinatorial libraries of nucleic acids, proteins or peptides
- G16B35/20—Screening of libraries
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
- G16B5/00—ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
- G16B5/20—Probabilistic models
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
- G16C20/60—In silico combinatorial chemistry
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2500/00—Screening for compounds of potential therapeutic value
- G01N2500/02—Screening involving studying the effect of compounds C on the interaction between interacting molecules A and B (e.g. A = enzyme and B = substrate for A, or A = receptor and B = ligand for the receptor)
Definitions
- the present invention relates to a method for identifying unknown molecular interactions within biological networks based on representations of molecules as sets of conserved features.
- molecules include but are not limited to proteins and nucleic acid molecules which can be represented as collections of conserved features, such as domains and motifs in proteins.
- the method of the invention comprises computing the attraction probabilities between molecules followed by calculation of the probability of a biological network.
- the method of the invention can be applied across species, where interaction data from one, or several species, can be used to infer molecular interactions between molecules within or between organisms
- the method of the present invention may be used to identify molecular interactions which can serve as drug screening targets.
- Markov Model methods for comparison of a set of aligned sequences that usually represent a protein motif or domain with a database e.g., Krogh et al., 1994, J. Mol.
- the present invention relates to a method for identifying unknown molecular interactions within biological networks based on the representations of molecules as collections of features, where each feature is responsible for a specific interaction with another feature.
- the invention is described herein for protein interactions, however the method of the invention can also be used to identify additional types of molecular interactions.
- a method is provided for identifying unknown molecular interactions within biological networks based on the representation of proteins as collections of conserved domains and motifs, where each domain is responsible for a specific interaction with another domain. By characterizing the frequency with which specific domain-domain interactions occur within known protein interactions, the method of the invention permits the assignment of a probability to an arbitrary interaction between any two proteins with defined domains.
- Domain interaction data may be complemented with information on the topology of a biological network and is inco ⁇ orated into the method by assigning greater probabilities to networks displaying more biologically realistic topologies.
- Markov chain Monte Carlo techniques can be utilized for the prediction of posterior probabilities of intervention between a set of proteins, allowing its application to large data sets.
- the method of the invention can be applied across species, where interaction data from one, or several species, can be used to infer interactions between proteins.
- the method is can be analogously applied to other molecular data such as nucleic acid molecules including DNA and RNA molecules.
- Figure 2 The number of domains per protein does not determine network connectivity.
- FIG. 4 Scale- free (self-similarity) properties of a common cauliflower plant: it is virtually impossible to determine whether one is looking at a photograph of a complete vegetable or its part, unless an additional scale-dependent object (a match) is added.
- A Complete vegetable
- B A small segment of the same vegetable, ( Small part of the segment shown in figure (B). The same match was used in all three photographs for providing a sense of scale for otherwise scale-free structure. The idea originated from the book written by Peitgen and colleagues (Peitgen et al.,1992 in Chaos and Fractals: New Frontiers of Science, New York, Springer- Verlag). The photographs were obtained by direct scanning of the objects with Compaq S 4 100 scanner.
- Figure 5 Log of network likelihood based only on network topology.
- A, B Represent, respectively, a highly (overly) connected and minimally connected network.
- C Depicts a more realistic (optimum) configuration.
- D Shows a network with the same number of edges as (Q but with a less-favorable arrangement of incoming and outgoing edges. Networks with less-negative Log scores are more likely. Networks were created with Cutenet (Koike and Rzhetsky, 2000, Gene 259: 235-244).
- Vertices are 1- transcription factor BAS1 (gi
- BAS1 gi
- 2-oxoglutarate dehydrogenase precursor gi
- Edge probabilities for the network based on domain-domain attraction probabilities alone b, Posterior probabilities of all edges of the network after 10 9 iterations of MCMC simulation. Red and blue colors represent probability above and below, respectively, the mean of all edge probabilities (in white). Note that only values at vertex intersections (+) have meaning-areas in between are interpolated and merely help to show gradients. Edges known to exist from the original data are ⁇ (3, 2), (4, 5), (5,4), (7,6), (9, 8), (9, 9), (10, 1), and (11,3) ⁇ . The percentage of rejected edges in MCMC computation was 82%. Figure 8. Prediction of interactions among 10 proteins that are involved in the human apoptosis pathway. Only probabilities based on domain- domain interactions alone are shown.
- Figure 9 Known and predicted edges. Known edges are shown as open circles, while predicted edges are displayed as an "x”. Figure 10. Network posterior probabilities.
- FIG. 12 Probabilities of interactions between features. 12 A. Tuple- Tuple. 12B Pfam domain-domain. 12C. Feature vector. 12D. Tuple-Pfam.
- Bio networks comprise proteins, nucleic acids, and small molecules as primary interacting elements. Functional areas that provide the ability for one molecule to interact with another are generally referred to as domains or motifs. For example, subsequences of DNA where specific proteins bind are one class of domain, as are the amino acid subsequence responsible for binding activity within the protein. Since genes are passive carriers of information, and because there are relatively few enzymatic or structural RNA molecules, the majority of important biological functions are carried out by proteins. Interactions between proteins are of particular interest, as they are responsible for the majority of "active" biological function. To date, protein-protein interactions are also the predominant type of interaction with significant quantities of supporting experimental data sets.
- proteins Being linear sequences of amino acids at the level of primary structure, at the functional level, proteins can be broken down into segments that correspond to functional domains or conserved motifs. Like amino acids, these domains are discrete "letters,” combinations of which give rise to the diversity of protein form and function.
- the existence of a network connection between proteins is a function of the domain composition of each.
- non-protein network nodes are treated as single-domain proteins. Moving along a network pathway, a domain of an upstream protein may favor interaction with a domain of a downstream protein.
- interaction may also represent more general relationships between domains, e.g. information flow.
- the method of the present invention is based on the assumption that once a given pair of interactions has proven effective, nature will tend to re-use it in other networks within the same organism, as well as in other organisms. Thus, the method is based on quantifying, from data taken from known networks, the frequency with which a domain in one protein is observed immediately upstream or downstream of domains in another protein. This information is then used to infer the probability of unknown interactions.
- the present invention relates to a method for identifying unknown molecular interactions within biological networks based on the representation of proteins as collections of conserved domains and motifs, where each domain is responsible for a specific interaction with another domain.
- the method assigns probabilities to all possible networks found from a fixed number of vertices and provides each network with a probability value in such a way that networks having more features typical of real networks have higher probabilities.
- Each vertex of the network is composed of one or more domains or motifs, which are identified through comparison with existing databases of protein domains (e.g. Pfam (Bateman et al., 2000, Nucleic Acids Res. 28:263-6).
- the frequency of separate occurrence of domains d m and d travers in two connected vertices of a known network is used to infer probabilities of "attraction" p(d m ,d n ),i.e., that an oriented edge will be found between these domains. As described in detail below, these probabilities are used to determine the probability of individual protein-protein interactions.
- the method comprises two independent stochastic steps, and the probability of an individual network emerges as a product of the probabilities associated with these two steps.
- every pair of proteins i and j may be connected to each other with an "attraction" probability p ⁇ -, or not connected with probability (1 -p, j ).
- this process is performed by a machine that, for every pair of vertices, tosses imaginary biased coins, each coin specific to a particular pair of proteins. If it is heads, an edge between the two vertices is formed; if it is tails, it does not.
- the coin is biased by prior information about the domains in each of the vertices, leading to some edges having a probability greater than 0.5 (attraction) and some edges less than 0.5 (repulsion).
- a network with ⁇ V ⁇ vertices there are 2' ⁇ possible networks with oriented edges.
- the probability of a single network with the particular edge set E is defined as:
- networks are sorted into a finite number of bins, each corresponding to a particular "network topology.”
- network topology is defined as the particular distribution of edges coming into and out of each vertex of the network.
- each bin represents a collection of networks with identical topologies.
- the number of incoming edges, or indegree, of a vertex in an oriented graph is the number of oriented edges that end at this vertex.
- the outdegree of a vertex is the number of oriented edges that start at this vertex.
- the upstream protein has a single outgoing edge while the downstream protein possess a single incoming edge.
- the number of vertices that have outdegree zero, no° ul , one, n ° ut , two, « "', and so on to n N 0Ut (where the subscript indicates the number of edges, and Nis the total number of vertices in the graph) is computed.
- the numbers of vertices with indegrees 0, 1,2,.. .N can be computed.
- Into one bin all networks with identical sets ⁇ n x ⁇ n ⁇ and ⁇ « "' ⁇ are added. For each bin a sampling probability, P( ⁇ n x '" ⁇ , ⁇ « "' ⁇ ) is defined that is computed as the product
- Each individual edge has a probability associated with it, and thus so does the complete set of edges that make up a given network P(E).
- a network topology is automatically defined when given this same set of edges E.
- the probability of this particular topology is determined separately and may or may not be biologically realistic. It is not necessary to distinguish between (a) and (b) above because topologically they are identical. They are not identical, however, at the level of individual edges.
- Each of these edges in (a) and (b) will have a different probability associated with them, with presumably one version of (a) or (b) being the correct, and thus most favorable one.
- To verify that the product of the former two stochastic steps would give the probability of sampling any given network the following equation may be utilized:
- Proteins are viewed as "collections of domains" where each individual pair of domains, d m and d n , has a probability of getting attracted, p(d m , d n ). If p(d m , d n ) > 0.5, the domains "attract” each other, while for p(d m , d n ) ⁇ 0.5, the domains "repel" each other.
- edge probability is reasonable insofar as the number of edges going into or out of a vertex is not correlated with the number of distinct domains in either of the interacting proteins.
- Interactions between proteins published in the research literature have strikingly different reliabilities. This is in part due to the fact that it is uncommon to publish the negative results of an experiment. As a result, the presence of an interaction between proteins is usually backed by multiple experiments while the absence of interaction may correspond to a failed experiment or just the absence of experiments at all (the only exclusion from this observation is exhaustive two-hybrid screening where all results, both positive and negative, are reported).
- ⁇ is a positive real-valued pseudocount
- k mn is the number of edges in the training set that contain at least one domain d m at the vertex of edge origin and at least one domain dn at the vertex of edge destination
- k m is the number of distinct vertices that contain at least one domain d m
- k cr is the number of distinct vertices that contain at least one domain d d
- ktile is the number of distinct vertices that contain at least one domain dministere.
- this formulation assigns probabilities greater than 50% to edges that have known connections, and probabilities equal to 50% to edges that have no known connections.
- the probability of an edge forming between any two domains is exactly 50%, which in turn leads to a probability of 50% for forming an edge between any two proteins, regardless of the number of domains in each of them.
- all networks can be assigned a non-zero probability.
- the current methodology specifically, Equation 6
- the probability range could be expanded from 0 - 1, the compressed scale of 0.5 - 1 does not affect the results.
- the method may be modified, combining the collection of appropriate data (e.g. negative experimental results, appropriate 2-hybrid data, etc.), with the range of 0 - 0.5 for modeling "repulsive" effects between domains.
- the method of the present invention assigns a probability to every possible network with ⁇ V ⁇ vertices. This probability is based on both local and global network properties. At the local level, the probability of a vertex having an interaction with another vertex is dependent on the domain composition of each. If, as previously determined by training data, the set of domains in one protein are likely to be attracted to those of another protein, the probability of an edge existing between the two vertices increases to a value greater than 0.5. If no information is available about the likelihood of interaction for the set of domains contained in both upstream and downstream vertices, the probability of an edge forming between the two is taken to be 0.5.
- the probability of the network (based solely on local properties) is modified based on how well it represents real biological networks. For example, networks with topologies (distribution of incoming and outgoing edges per vertex) that are more biologically realistic are given higher likelihoods. The probability of any given network is the product of both the local and global probabilities.
- a bootstrap procedure was then ran where for 200 iterations, where 30 vertices were randomly removed from the network and the value of ⁇ and 95% confidence intervals were determined for each. After this was completed, 60 vertices were removed and the process repeated. This was repeated until the final 200 iterations with 113 total vertices in the network.
- the effect of vertex removal on ⁇ is shown in Figure 6 (mean of ⁇ and 95% confidence intervals displayed), and shows that this network is remarkably scale-invariant. This implies that knowledge of the topology of a small part of a network should provide a reliable means of estimating the complete network's topology.
- Cross-validation was used to determine the effectiveness of the model in predicting the overall network configuration.
- Cross- validation is a general technique for evaluating the efficiency (and hence, validity) of statistical algorithms. It typically involves dividing a data set into two disjoint subsets, one of which is used for training, with the other being used for model validation.
- the jackknife version of cross-validation was used, where the training set consists of the complete network minus a single, specified edge.
- the likelihood of the complete graph (the model validation set of data) was compared to that of the complete graph minus one edge. If the likelihood of the full network was greater than the likelihood of the reduced network, the edge was considered to be positively predicted. This step was performed iteratively until all edges had been considered.
- Markov chain Monte Carlo For nearly all species, many interactions within a biological pathway are currently unknown. Since the method of the present invention permits computation of the probability for any possible arrangement of edges that connect a set of vertices, a Markov chain Monte Carlo (MCMC) simulation approach (Gilks et al, (editors) in Markov chain Monte Carlo in Practice. Chapman & Hall, New York; Hastings ,1970 Biometrika 57:97-109) can be implemented; which allows computation of posterior probabilities for all edges while effectively sampling from the astronomically large number of possible networks.
- MCMC Markov chain Monte Carlo
- a reversible-jump methodology (Green, 1995 Biometrika 57 82:711- 732) typical for Bayesian model selection was implemented, treating different networks as alternative statistical models.
- a uniform prior distribution was chosen over all networks, because, without additional information, there is no reason to prefer one network over another.
- the algorithm either adds or removes, with equal probability, a defined number of edges. Edges to be added or deleted are respectively sampled from the pool of edges that are included or excluded from the current network, with the probability of selecting any given edge dependent on only the number of edges from which to choose. Adding or removing edges in this manner, the system jumps from network X to a new network Y.
- the proposed new state Fis sampled from the proposal distribution, q(b ⁇ a).
- the new network 7 is then accepted with probability
- L(.) is the likelihood of the given network. If the proposed new state is accepted, then network Y becomes the current network; otherwise, the old network X remains as the current model.
- the stochastic process moves through the space of possible networks, on average keeping each edge in an on or off state in proportion to the posterior probabilities of this edge being present or absent in the correct network.
- a group of 11 yeast proteins known to interact with at least one other member of the group was selected, and an attempt was made to predict edges (Figure 6).
- the probabilities of a given edge, based on domain- domain interactions alone, are shown in part a. Note that all edges except (7,1) (x- axis, y-axis) are found in the original data.
- edges (7, 1) and (10, 1) are not supported with high confidence due to their low domain-domain interaction probabilities and to the influence of the edge distributions.
- regions of low probability e.g. the vicinity of (4, 8)
- regions of low probability e.g. the vicinity of (4, 8)
- regions of low probability e.g. the vicinity of (4, 8)
- proteins that already have a high-probability edge
- addition of a second edges is unlikely.
- the nonsymmetrical pattern is due to differences between the outgoing and incoming edge distributions. Although they are easily differentiated from unlikely edges, all likely edges have relatively low posterior probabilities.
- MCMC method For very small systems, a significant amount of information can be gained simply from looking at the edge probabilities between a given set of vertices, with very little additional information coming from topology information.
- use of the MCMC method described here should be particularly valuable for the prediction of large networks, where large amounts of protein interaction data with complicated domain architectures (such as those of higher organisms), and a computationally intensive number of network topologies, is the norm.
- the present invention permits both characterization and prediction of both known and unknown protein interactions within a given species, and potentially, across species.
- Markov chain Monte Carlo techniques described earlier provide a computationally feasible way to calculate the posterior probability of a network given data as:
- the method of the present invention could significantly reduce the number of required experiments by identifying a few most likely hypotheses.
- Such experimental analysis is itself an empirical way of validating the model, and can facilitate likewise design experiments for validation.
- Improvements could include additional interaction data and the introduction of more domains for assignment to protein segments.
- the method may be further enhanced by allowing the introduction of repulsion effects, which are implemented by allowing probabilities of less than 0.5 for domain -domain interactions. This information can be gathered from experiments (past and future) as well as from experts in the field.
- the creation of pseudo-domains for characterizing non-protein substances and small molecules would allow their analysis within the network.
- various molecular parameters e.g.
- Proteins used in this analysis are: 1) ANT2, 2) APP (695), 3) B-CAT, 4) BAG3, 5) BAK, 6) Bax-beta, 7) Bcl-xL, 8) BCL2A1, 9) Be 12- alpha, 10) Calsenilen, 11) CAV1, 12) CHIP, 13) CD3, 14) D-CAT, 15) DRAL, 16) FLN1, 17) FLNB, 18) GAPCenA, 19) GDI1, 20) GDI2, 21) GGTB, 22) GTPBP1, 23) HSPA4, 24) HSPA8, 25) KSR1, 26) MCL1, 27) MRJ, 28) PSAP, 29) PKP4, 30) PLCG1, 31) PS1 (467), 32) PS2 (448), 33) QM, 34) RAB11A, 35) RAB3A, 36) RAB5A, 37) RAB6, 38) RAB6KIFL, 39) TF, 40) TTC
- L(.) is the likelihood of the network. If the proposed state is accepted, it becomes the current state. This method thus samples networks from the space of all possible networks while keeping each edge occupied, or unoccupied over time, in proportion to its posterior probability.
- FIG. 10A The posterior distribution generated from approximately 10 7 samples is shown in Figure 10 A-B.
- 10A it can be seen that a few edges are readily apparent; rising well above the surrounding background. The two tallest peaks are of the HSPA8-MRJ interaction. Edges such as these show up rapidly in the simulations, while low-probability edges can take considerably greater amounts of sampling to distinguish them from background.
- Figure 10B shows the posterior probabilities for each edge of the network.
- the lower probability (darker) "lines” running horizontally at vertices 20 and 27 and vertically along vertex 27 show the influence of the nonsymmetrical edge distributions. For example, since vertex 27 has a high probability connection, the edge distribution tends to suppress the addition of new edges to the same vertex.
- any vertex can have multiple incoming and outgoing edges, however due to the scale- free property of these networks, highly connected vertices are relatively rare.
- Figure 11 A-B the negative multinomial with different parameters P t is shown, while Figure 11C shows a multinomial distribution. It can be seen that by increasing E, it is possible to increase the variance of the distribution while keeping the expected value identical to the multinomial distribution shown in part c. However, while one can match the expected value, one can only generate a variance that is greater than, but not equal to, the multinomial's. This is because a negative binomal distribution tends to a Poisson distribution as the variance decreases, and the Poisson distribution has typically larger variance than a multinomial distribution with the same mean.
- One means for improving the method of the invention is to implement "repulsive" interactions between domains. This can be achieved by assigning domain-domain interaction probabilities of ⁇ 0.5 to interactions that are never present. While requiring careful normalization and balancing with “attractive" probabilities, this feature should provide enhanced resolution of predicted interactions (bigger peaks and deeper valleys in the posterior plots). While having its own set of favorable and unfavorable properties, two-hybrid data should prove particularly valuable for this approach.
- the lowest level corresponds to four adjacent amino acids observed in a running window which spanned the entire length of the protein.
- Table 1 Mapping of amino acids to group numbers.
- Feature class describes the general properties of the group.
- the intermediate-scale feature consisted often adjacent amino acids, again taken from a running window spanning the entire protein.
- six particular traits were looked at including positive charge, negative charge, hydrophobicity, amphipathy, proline-richness, and serine-richness.
- Each ten-mer was analyzed for the density of each of these traits, and a six-element feature vector generated containing the strength of each of the six traits. From this feature vector, the two best-represented traits were chosen to represent this particular ten-mer, with vectors containing multiple identical top traits having the top two features chosen at alphabetical order.
- this representation is not specific as to the order of amino acids within a 10-mer, but rather first describes the density of each of a set of traits and then assigns only the best two to the ten-mer.
- the scales and features described here could easily have been chosen in a variety of different biologically relevant ways. For instance, the intermediate scale could be redefined, for example, with specific degrees of hydrophobicity, surface tension, etc. determined for each amino acid as well as for each ten-mer. While the features chosen are not necessarily optimal the current implementation provides sufficient detail for this level of analysis.
- Probabilities of interactions were assigned as follows. As described earlier, it was assumed that some set of features is responsible for the observed interaction within each protein in an interacting pair. To determine the specific pairs of features giving rise to the interaction, and then translate this information into a prediction for proteins for which there is no information, training data consisting of a large number of known interactions between proteins with known sequences was used. Using this training data an attempt was made to infer whether an interaction exists between a set of proteins based upon analyzing the feature set of these potentially interacting proteins. Note that the assignment of protein-protein interaction probabilities is related to previous work, but now makes use of both positive (existing interactions) and negative (absence of interactions) information from the training data, while other versions of the approach described herein discarded the negative information.
- the model was set so that, in the absence of interaction data, any pair of features will interact with a probability of 0.5.
- the probability of feature-feature interaction less than 0.5 indicates that the features "repel,” while probability greater than 0.5 indicates that the features "attract.”.
- the probability of a pair of features (d t ⁇ dj) interacting with one another is estimated in the following way
- n * is the number of times feature i is seen in an interaction with feature y (the number of times two interacting proteins have features andy, respectively)
- n j is the number of times feature i is not seen in an interaction with feature y (the number of times two non-interacting proteins display features i andj, respectively).
- ⁇ is a parameter that reflects non-uniform distribution of different types of data, negative and positive, and has direct relation to concept called "boosting" in computer science.
- y is the probability of protein i interacting with protein Since the uninformative prior probability of any particular interaction is 0.5, the total information content of a set of protein interactions is defined as
- a major benefit of the multi-scale protein characterization described here is the ability to use all possible interaction data for training.
- characterization of proteins with Pfam domains alone meant that the majority of protein interactions could not be defined; a drawback that has affected other models.
- using less stringent E-value cutoffs could perhaps help to increase the number of domains in the network without producing excessive noise, even relatively relaxed cutoffs provide only 60-70% coverage at best.
- the vector of features (stretches of positive charge, negative charge, hydrophobic, aliphatic, proline, and serine residues) was chosen primarily for proof of concept, with features thought to be of general utility and with some predictive capability. Windows greater than 10 amino acids could obviously be used, and could perhaps be useful in detecting large-scale structures important in establishing interactions between proteins.
- a related aspect of this approach is the hope that the characterization of features at multiple scales could potentially provide the ability to observe small-scale effects. For instance, it may be possible to observe substitution events that cause a loss or gain of function - generating a novel interaction or deleting an existing one. Characterization with small-scale features would be required for this type of analysis. The determination of which particular types features in the vector were most informative is currently being evaluated.
Abstract
Description
Claims
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2002564586A JP2005503535A (en) | 2001-02-09 | 2002-02-11 | Method for predicting molecular interaction networks |
EP02706238A EP1360483A4 (en) | 2001-02-09 | 2002-02-11 | A method for the prediction of molecular interaction networks |
CA002437878A CA2437878A1 (en) | 2001-02-09 | 2002-02-11 | A method for the prediction of molecular interaction networks |
Applications Claiming Priority (8)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US26797001P | 2001-02-09 | 2001-02-09 | |
US60/267,970 | 2001-02-09 | ||
US26922001P | 2001-02-16 | 2001-02-16 | |
US60/269,220 | 2001-02-16 | ||
US32360001P | 2001-09-20 | 2001-09-20 | |
US32359901P | 2001-09-20 | 2001-09-20 | |
US60/323,600 | 2001-09-20 | ||
US60/323,599 | 2001-09-20 |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2002065119A1 WO2002065119A1 (en) | 2002-08-22 |
WO2002065119A9 true WO2002065119A9 (en) | 2004-01-15 |
Family
ID=27500923
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2002/004028 WO2002065119A1 (en) | 2001-02-09 | 2002-02-11 | A method for the prediction of molecular interaction networks |
Country Status (5)
Country | Link |
---|---|
US (1) | US20030068610A1 (en) |
EP (2) | EP2051177A1 (en) |
JP (1) | JP2005503535A (en) |
CA (1) | CA2437878A1 (en) |
WO (1) | WO2002065119A1 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9111026B1 (en) | 2003-08-22 | 2015-08-18 | Dennis Sunga Fernandez | Integrated biosensor and simulation system for diagnosis and therapy |
US9384323B1 (en) | 2002-10-18 | 2016-07-05 | Dennis S. Fernandez | Pharmaco-genomic mutation labeling |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7415359B2 (en) | 2001-11-02 | 2008-08-19 | Gene Network Sciences, Inc. | Methods and systems for the identification of components of mammalian biochemical networks as targets for therapeutic agents |
WO2003040992A1 (en) * | 2001-11-02 | 2003-05-15 | Gene Network Sciences, Inc. | Methods and systems for the identification of components of mammalian biochemical networks as targets for therapeutic agents |
US7451065B2 (en) * | 2002-03-11 | 2008-11-11 | International Business Machines Corporation | Method for constructing segmentation-based predictive models |
US20040236515A1 (en) * | 2003-05-20 | 2004-11-25 | General Electric Company | System, method and computer product for predicting protein- protein interactions |
US20070174019A1 (en) * | 2003-08-14 | 2007-07-26 | Aditya Vailaya | Network-based approaches to identifying significant molecules based on high-throughput data analysis |
US7580570B2 (en) * | 2003-12-09 | 2009-08-25 | Microsoft Corporation | Accuracy model for recognition signal processing engines |
US20050154535A1 (en) * | 2004-01-09 | 2005-07-14 | Genstruct, Inc. | Method, system and apparatus for assembling and using biological knowledge |
US20080261820A1 (en) * | 2005-08-01 | 2008-10-23 | Mount Sinai School Of Medicine Of New York University | Methods to Analyze Biological Networks |
JP5024583B2 (en) * | 2005-09-14 | 2012-09-12 | ソニー株式会社 | Information processing apparatus and information processing method, information processing system, program, and recording medium |
JP5667822B2 (en) | 2010-09-21 | 2015-02-12 | 株式会社日立製作所 | Parts mounting structure in the windmill tower |
CN117198426B (en) * | 2023-11-06 | 2024-01-30 | 武汉纺织大学 | Multi-scale medicine-medicine response interpretable prediction method and system |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU6271394A (en) * | 1993-02-23 | 1994-09-14 | General Hospital Corporation, The | A computer system and method for measuring an analyte concentration with an affinity assay |
US5604100A (en) * | 1995-07-19 | 1997-02-18 | Perlin; Mark W. | Method and system for sequencing genomes |
US6132969A (en) * | 1998-06-19 | 2000-10-17 | Rosetta Inpharmatics, Inc. | Methods for testing biological network models |
US6203987B1 (en) * | 1998-10-27 | 2001-03-20 | Rosetta Inpharmatics, Inc. | Methods for using co-regulated genesets to enhance detection and classification of gene expression patterns |
US6772069B1 (en) * | 1999-01-29 | 2004-08-03 | University Of California, Los Angeles | Determining protein function and interaction from genome analysis |
AU2001278089A1 (en) * | 2000-07-31 | 2002-02-13 | Agilix Corporation | Visualization and manipulation of biomolecular relationships using graph operators |
US6594587B2 (en) * | 2000-12-20 | 2003-07-15 | Monsanto Technology Llc | Method for analyzing biological elements |
-
2002
- 2002-02-11 JP JP2002564586A patent/JP2005503535A/en active Pending
- 2002-02-11 CA CA002437878A patent/CA2437878A1/en not_active Abandoned
- 2002-02-11 US US10/073,463 patent/US20030068610A1/en not_active Abandoned
- 2002-02-11 EP EP08020358A patent/EP2051177A1/en not_active Withdrawn
- 2002-02-11 WO PCT/US2002/004028 patent/WO2002065119A1/en active Application Filing
- 2002-02-11 EP EP02706238A patent/EP1360483A4/en not_active Ceased
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9384323B1 (en) | 2002-10-18 | 2016-07-05 | Dennis S. Fernandez | Pharmaco-genomic mutation labeling |
US9454639B1 (en) | 2002-10-18 | 2016-09-27 | Dennis Fernandez | Pharmaco-genomic mutation labeling |
US9582637B1 (en) | 2002-10-18 | 2017-02-28 | Dennis Sunga Fernandez | Pharmaco-genomic mutation labeling |
US9111026B1 (en) | 2003-08-22 | 2015-08-18 | Dennis Sunga Fernandez | Integrated biosensor and simulation system for diagnosis and therapy |
US9110836B1 (en) | 2003-08-22 | 2015-08-18 | Dennis Sunga Fernandez | Integrated biosensor and simulation system for diagnosis and therapy |
Also Published As
Publication number | Publication date |
---|---|
EP1360483A1 (en) | 2003-11-12 |
JP2005503535A (en) | 2005-02-03 |
EP1360483A4 (en) | 2008-03-05 |
CA2437878A1 (en) | 2002-08-22 |
WO2002065119A1 (en) | 2002-08-22 |
US20030068610A1 (en) | 2003-04-10 |
EP2051177A1 (en) | 2009-04-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Aittokallio et al. | Graph-based methods for analysing networks in cell biology | |
Sharon et al. | A feature-based approach to modeling protein–DNA interactions | |
Gomez et al. | Towards the prediction of complete protein–protein interaction networks | |
US20030068610A1 (en) | Method for the prediction of molecular interaction networks | |
Ferrarini et al. | A more efficient search strategy for aging genes based on connectivity | |
Solis | Reduced alphabet of prebiotic amino acids optimally encodes the conformational space of diverse extant protein folds | |
Fraser et al. | What can be learned by scanning the genome for molecular convergence in wild populations? | |
Iossifov et al. | Probabilistic inference of molecular networks from noisy data sources | |
Godden et al. | Recursive median partitioning for virtual screening of large databases | |
Carter | Inferring network interactions within a cell | |
Deng et al. | EXAMINE: A computational approach to reconstructing gene regulatory networks | |
Altman et al. | Defining bioinformatics and structural bioinformatics | |
Sharma et al. | RBPSpot: Learning on appropriate contextual information for RBP binding sites discovery | |
Hu et al. | A novel method for discovering local spatial clusters of genomic regions with functional relationships from DNA contact maps | |
AU2002240340A1 (en) | A method for the prediction of molecular interaction networks | |
Golden et al. | Evolutionary analyses of base-pairing interactions in DNA and RNA secondary structures | |
Kinghorn et al. | Selective phenome growth adapted model: A novel landscape to represent aptamer ligand binding | |
Lihu et al. | De novo motif prediction using the fireworks algorithm | |
Belda et al. | Evolutionary computation and multimodal search: A good combination to tackle molecular diversity in the field of peptide design | |
Wrzeszczynski et al. | Cataloging proteins in cell cycle control | |
Basak | Chemoinformatics and bioinformatics by discrete mathematics and numbers: an adventure from small data to the realm of emerging big data | |
Rout et al. | Protein secondary structure prediction of PDB 4HU7 using Genetic Algorithm (GA) | |
Das | Implementation of machine learning technique to predict essential genes of Saccharomyces cerevisiae | |
Mahdevar et al. | Transcription factor binding sites detection by using alignment-based approach | |
Gesell et al. | Phylogeny and evolution of RNA structure |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ OM PH PL PT RO RU SD SE SG SI SK SL TJ TM TN TR TT TZ UA UG UZ VN YU ZA ZM ZW |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2002564586 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2437878 Country of ref document: CA |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2002240340 Country of ref document: AU |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2002706238 Country of ref document: EP |
|
WWP | Wipo information: published in national office |
Ref document number: 2002706238 Country of ref document: EP |
|
REG | Reference to national code |
Ref country code: DE Ref legal event code: 8642 |
|
COP | Corrected version of pamphlet |
Free format text: PAGES 1/12-12/12, DRAWINGS, REPLACED BY NEW PAGES 1/13-13/13; DUE TO LATE TRANSMITTAL BY THE RECEIVING OFFICE |