Patent Application
20080275683
1. Field of the Invention
The present invention generally relates to computing techniques for modeling physical interactions between two substances at a molecular level. More specifically, the present invention relates to computing techniques that use a kd-tree and envelope to improve identification of nearest atoms as part of a computational simulation.
2. Description of the Related Art
Powerful computers may be designed as highly parallel systems where the processing activity of hundreds, if not thousands, of processors (CPUs) are coordinated to perform computing tasks. These systems are highly useful for a broad variety of applications including, financial modeling, hydrodynamics, quantum chemistry and mechanics, astronomy, weather modeling and prediction, geological modeling, prime number factoring, image processing (e.g., CGI animations and rendering), to name but a few examples.
One family of parallel computing systems has been (and continues to be) developed by International Business Machines (IBM) under the name Blue Gene®. The Blue Gene/L architecture provides a scalable, parallel computer that may be configured with a maximum of 65,536 (216) compute nodes. Each compute node includes a single application specific integrated circuit (ASIC) with 2 CPU's and memory. The Blue Gene/L architecture has been successful and on Oct. 27, 2005, IBM announced that a Blue Gene/L system had reached an operational speed of 280.6 teraflops (280.6 trillion floating-point operations per second), making it the fastest computer in the world at that time. Further, as of June 2005, Blue Gene/L installations at various sites world-wide were among five out of the ten top most powerful computers in the world.
IBM is currently developing a successor to the Blue Gene/L system, named Blue Gene/P. Blue Gene/P is expected to be the first computer system to operate at a sustained 1 petaflops (1 quadrillion floating-point operations per second). Like the Blue Gene/L system, the Blue Gene/P system is scalable with a projected maximum of 73,728 compute nodes. Each compute node in Blue Gene/P is projected to include a single application specific integrated circuit (ASIC) with 4 CPU's and memory. A complete Blue Gene/P system is projected to include 72 racks with 32 node boards per rack.
In addition to the Blue Gene architecture developed by IBM, other highly parallel computer systems have been (and are being) developed. For example, a Beowulf cluster may be built from a collection of commodity off-the-shelf personal computers. In a Beowulf cluster, individual systems are connected using local area network technology (e.g., Ethernet) and system software is used to execute programs written for parallel processing on the cluster of individual systems.
As stated, these, and other, parallel systems are often used to perform simulations of molecular systems. One such type of simulation is used to determine whether one compound (referred to as a ligand) will bind to another compound (referred to as a receptor). These simulations are expected to lead to discoveries of new useful drugs and new medical treatment methods. For example, these simulations may be performed to identify a compound that will deliver a particular therapeutic substance to a particular location on a particular protein (e.g., a compound that will target a particular site on the surface of a cancerous cell).
In order to determine whether a compound is likely to bind with a receptor, multiple iterations of a simulation are usually performed to account for the various conformations in which the ligand and receptor may encounter one another. That is, the simulation may evaluate a very large number of possible conformations in which the ligand may bind with the receptor. For each conformation, the simulation may be configured to determine whether the conformation is possible (i.e., likely to occur) and, if so, whether the ligand will bind with the receptor.
Given the nature of this (and other similar) problems, parallel computing has emerged as the preferred way to perform these simulations because a very large number of conformations can be tested simultaneously on the compute nodes of a parallel system. Of course, molecular simulations may be performed on more conventional computer systems; they just take significantly longer to perform.
As stated, these types of molecular simulations may first evaluate whether a given receptor/ligand conformation is physically possible. For example, a conformation may position an atom from the ligand at a point too close to an atom in the receptor. That is, the conformation may specify a state for the ligand and receptor that cannot (or is highly unlikely) to occur in the real world, based on our understanding of quantum mechanics. If the atoms are too close, then the results of any free energy calculations based on that conformation are unlikely to produce any meaningful data.
Traditionally, a brute force method is used to ensure that none of the ligand atoms are too close to the receptor atoms. That is, the simulation checks all n atoms of the ligand against all m atoms of the receptor. This leads to a runtime requirement of m*n comparisons for a single conformation. And recall, this process is usually performed for many thousands of different confirmations between a ligand and receptor, and performed for hundreds of ligands (if not thousands or more). As a result, the performance cost of performing an n*m compassion for each conformation is magnified many times. Commonly assigned patent application “Envelope Technique for Exclusion of Atoms in an Hbond Check” (atty. docket ROC-9-2006-0210) describes an improvement to this technique where only atoms within a defined region of space are selected for comparison. However, each of the m′ atoms of the receptor within the envelope are still compared against each of the n′ atoms of the ligand within the envelope, leading to n′*m′ comparisons being performed for a single conformation.
Accordingly, as the foregoing illustrates, there remains a need for other techniques used to perform these (and other similar) types of molecular modeling simulations.
Embodiments of the present invention provide a technique for using a kd-tree and envelope to improve identification of nearest atoms as part of a computational simulation.
One embodiment of the invention provides a method of performing a computational simulation. The method generally includes selecting a conformation for a first molecule and a second molecule to simulate. The conformation specifies a set of atoms in the first molecule, a set of atoms in the second molecule, and specifies a position of the first and second molecule, relative to one another. The method also includes determining a region of space for an envelope surrounding the set of atoms in the second molecule and parsing a first generated kd-tree based on atoms of the first molecule to identify which atoms of the first molecule are within the envelope surrounding the set of atoms in the second molecule. For each atom of the first molecule within the envelope surrounding the set of atoms in the second molecule, the method also includes parsing a second generated kd-tree based on the atoms of the second module to identify a corresponding nearest atom of the second molecule.
In a particular embodiment, the method also includes generating a first kd-tree based on atoms of the first molecule. The nodes of the first kd-tree store the positions of the atoms in the first molecule. The method may also include generating a second kd-tree based on atoms of the second molecule. The nodes of the second kd-tree store the positions of the atoms in the first molecule.
Another embodiment of the invention includes a computer-readable storage medium containing a program which, when executed, performs an operation for performing a computational simulation. The operation generally includes selecting a conformation for a first molecule and a second molecule to simulate. The conformation specifies a set of atoms in the first molecule, a set of atoms in the second molecule, and specifies a position of the first and second molecule, relative to one another. The operation also includes determining a region of space for an envelope surrounding the set of atoms in the second molecule and parsing a first generated kd-tree based on atoms of the first molecule to identify which atoms of the first molecule are within the envelope surrounding the set of atoms in the second molecule. For each atom of the first molecule within the envelope surrounding the set of atoms in the second molecule, the operation also includes, parsing a second generated kd-tree based on the atoms of the second module to identify a corresponding nearest atom of the second molecule.
Still another embodiment of the invention provides a computing device. The computing device generally includes a compute node having at least a processer and a memory and a simulation program, which when executed by the compute node, performs an operation. The operation generally includes selecting a conformation for a first molecule and a second molecule to simulate. The conformation specifies a set of atoms in the first molecule, a set of atoms in the second molecule, and specifies a position of the first and second molecule, relative to one another. The operation also includes determining a region of space for an envelope surrounding the set of atoms in the second molecule and parsing a first generated kd-tree based on atoms of the first molecule to identify which atoms of the first molecule are within the envelope surrounding the set of atoms in the second molecule. For each atom of the first molecule within the envelope surrounding the set of atoms in the second molecule, the operation also includes, parsing a second generated kd-tree based on the atoms of the second module to identify a corresponding nearest atom of the second molecule.
So that the manner in which the above recited features, advantages and objects of the present invention are attained and can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to the embodiments thereof which are illustrated in the appended drawings.
It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered limiting of its scope, for the invention may admit to other equally effective embodiments.
Embodiments of the invention provide a technique for improving the efficiency of a molecular modeling simulation. For example, embodiments of the invention may be used to reduce the number of comparisons performed in a simulation of binding affinity between a ligand and receptor. Because such a simulation is typically performed a very large number of times for different conformations of even a single ligand and receptor, the effect of reducing the number of comparisons for any individual conformation is highly leveraged and can provide a significant improvement to overall simulation performance.
In one embodiment, the positions of atoms in a receptor and ligand are represented using kd-trees. As is known, a kd-tree (short for k-dimensional tree) is a space-partitioning data structure for organizing points in a k-dimensional space. In the context of the present invention, the relative positions of each atom in a receptor and each atom in a ligand are stored as a node in a kd-tree. The simulation may also generate a spherical envelope that contains all atoms of the ligand. The sphere includes a center point and a radius, i.e., a point and a distance from that point defining the surface of the sphere. Techniques for generating the sphere are described in commonly assigned patent application “Envelope Technique for Exclusion of Atoms in an Hbond Check” (atty. docket ROC-9-2006-0210).
The simulation may be configured to parse the kd-tree representing the receptor to identify any receptor atoms that are within the spherical envelope (i.e., any atom of the receptor with a distance to the center of the sphere less than the radius of the sphere). Once identified, the atom of the ligand that is nearest to each atom in the receptor (within the envelope) may be identified. The comparison for being “too close” is then performed for the each atom of the receptor within the envelope and the corresponding nearest atom of the ligand. Thus, the number distance comparisons may be significantly reduced.
Although some additional overhead is incurred in generating and parsing the kd-trees, it is frequently more than offset by the reduction in the number of distance comparisons using a brute force approach between m atoms in the receptor and n atoms in the ligand (leading to m×n comparisons).
In the following, reference is made to embodiments of the invention. However, it should be understood that the invention is not limited to specific described embodiments. Instead, any combination of the following features and elements, whether related to different embodiments or not, is contemplated to implement and practice the invention. Furthermore, in various embodiments the invention provides numerous advantages over the prior art. However, although embodiments of the invention may achieve advantages over other possible solutions and/or over the prior art, whether or not a particular advantage is achieved by a given embodiment is not limiting of the invention. Thus, the following aspects, features, embodiments and advantages are merely illustrative and are not considered elements or limitations of the appended claims except where explicitly recited in a claim(s). Likewise, reference to “the invention” shall not be construed as a generalization of any inventive subject matter disclosed herein and shall not be considered to be an element or limitation of the appended claims except where explicitly recited in a claim(s).
One embodiment of the invention is implemented as a program product for use with a computer system. The program(s) of the program product defines functions of the embodiments (including the methods described herein) and can be contained on a variety of computer-readable media. Illustrative computer-readable media include, but are not limited to: (i) non-writable storage media (e.g., read-only memory devices within a computer such as CD-ROM or DVD-ROM disks readable by a CD- or DVD-ROM drive) on which information is permanently stored; (ii) writable storage media (e.g., floppy disks within a diskette drive or hard-disk drive) on which alterable information is stored. Other media include communications media through which information is conveyed to a computer, such as through a computer or telephone network, including wireless communications networks. The latter embodiment specifically includes transmitting information to/from the Internet and other networks. Such computer-readable media, when carrying computer-readable instructions that direct the functions of the present invention, represent embodiments of the present invention.
Embodiments of the invention are well suited for use with highly-parallel computer systems, such as the Blue Gene system developed by IBM. Accordingly,
As shown, computer system 100 includes a compute core 101 having a number of compute nodes arranged in a regular array or matrix, which perform the useful work performed by system 100. The operation of computer system 100, including compute core 101, may be controlled by control subsystem 102. Various additional processors in front-end nodes 103 may perform auxiliary data processing functions and file servers 104 provide an interface to data storage devices such as disk based storage 109A and 109B or other I/O (not shown). Functional network 105 provides the primary data communication path among compute core 101 and other system components. For example, data stored in storage devices attached to file servers 104 is loaded and stored to other system components through functional network 105.
Also as shown, compute core 101 includes I/O nodes 111A-C and compute nodes 112A-I. Compute nodes 112 provide the processing capacity of parallel system 100, and are configured to execute applications written for parallel processing. I/O nodes 111 handle I/O operations on behalf of compute nodes 112. Each I/O node 111 may include a processor and interface hardware that handles I/O operations for a set of q compute nodes 112, the I/O node and its respective set of q compute nodes are referred to as a Pset. Compute core 101 contains p Psets 11-5A-C, each including a single I/O node 111 and q compute nodes 112, for a total of p×q compute nodes 112. The product p×q can be very large. For example, in one implementation p=1024 (1 K) and q=64, for a total of 64K compute nodes.
In general, application programming code and other data input required by compute core 101 to execute user applications, as well as data output produced by the compute core 101, is communicated over functional network 105. The compute nodes within a Pset 115 communicate with the corresponding I/O node over a corresponding local I/O tree network 113A-C. The I/O nodes, in turn, are connected to functional network 105, over which they communicate with I/O devices attached to file servers 104, or with other system components. Thus, the local I/O tree networks 113 may be viewed logically as extensions of functional network 105, and like functional network 105 are used for data I/O, although they are physically separated from functional network 105.
Control subsystem 102 directs the operation of the compute nodes 112 in compute core 101. Control subsystem 102 is a computer that includes a processor (or processors) 121, internal memory 122, and local storage 125. An attached console 107 may be used by a system administrator or similar person. Control subsystem 102 may also include an internal database which maintains state information for the compute nodes in core 101, and an application which may be configured to, among other things, control the allocation of hardware in compute core 101, direct the loading of data on compute nodes 111, and perform diagnostic and maintenance functions.
Control subsystem 102 communicates control and state information with the nodes of compute core 101 over control system network 106. Network 106 is coupled to a set of hardware controllers 108A-C. Each hardware controller communicates with the nodes of a respective Pset 115 over a corresponding local hardware control network 114A-C. The hardware controllers 108 and local hardware control networks 114 are logically an extension of control system network 106, although physically separate.
In addition to control subsystem 102, front-end nodes 103 provide computer systems used to perform auxiliary functions which, for efficiency or otherwise, are best performed outside compute core 101. Functions which involve substantial I/O operations are generally performed in the front-end nodes. For example, interactive data input, application code editing, or other user interface functions are generally handled by front-end nodes 103, as is application code compilation. Front-end nodes 103 are connected to functional network 105 and may communicate with file servers 104. In one embodiment, compute nodes 112 are arranged logically in a three-dimensional torus, where each compute node may be identified using an x, y and z coordinate.
As described, functional network 105 may service many I/O nodes, and each I/O node is shared by multiple compute nodes 112. Thus, it is apparent that the I/O resources of parallel system 100 are relatively sparse when compared to computing resources. Although it is a general purpose computing machine, parallel system 100 is designed for maximum efficiency in applications which are computationally intense.
As shown in
Application code image 212 represents a copy of the application code being executed by compute node 112. Application code image 212 may include a copy of a computer program being executed by parallel system 100, but where the program is very large and complex, it may be subdivided into portions which are executed by different compute nodes 112. For example, Application code image 212 may include a sequence of instructions which, when executed by compute node 112, perform an operation for performing computational simulation. The simulation may include the use of a kd-tree and envelope to improve identification of nearest atoms as part of the computational simulation. In such a case, when performed essentially simultaneously by as many as 65,536 compute nodes of parallel system 100, a vast number of possible conformations between a ligand and a receptor may be evaluated. Memory 202 may also include a call-return stack 215 for storing the states of procedures which must be returned to, which is shown separate from application code image 202, although in may be considered part of application code state data.
As part of ongoing operations, application 212 may transmit messages from compute node 112 to other compute nodes in parallel system 100. For example, the high level MPI call of MPI_Send( ); may be used by application 312 to transmit a message from one compute node to another. On the other side of the communication, the receiving node may call use the MPI call MPI_Recieve( ); to receive and process the message. In context of the present invention, for example, a message may be sent from a control node to a compute node describing a conformation of a ligand and receptor to be evaluated by the receiving node. The receiving node may perform the simulation and then generate and transmit a message back regarding the results.
Other parallel systems also include a mechanism for transmitting messages between different compute nodes. For example, nodes in a Beowulf cluster may communicate using a using a high-speed Ethernet style network.
User node 302 may provide an interface to cluster 300. As such, user node 302 allows users to create, submit, and review the results of computing tasks submitted for execution to cluster 300. As shown, user node 302 is connected to head/gateway node 304. Head/gateway node 304 connects the user node 302 to the compute nodes 306. Compute nodes 306 provide the processing power of cluster 300. As is known, clusters are often built from racks of commonly available PC components. Thus, each node 306 may include one or more CPUs, memory, hard disk storage, a connection to high speed network switch 308, and other common PC components. Like the compute nodes 112 of parallel system 100, a compute node 306 of cluster 300 may execute a molecular modeling simulation evaluating a conformation of a ligand and a receptor.
Each node in a kd-tree includes a pointer to a node of a left branch and a pointer to a node of a right branch, each branch forming a kd-tree. Levels of the tree are split along successive k dimensions. For example, for a two-dimensional kd-tree, the tree may first be split using the x-dimension, then the y-dimension, then the x-dimension, then the y-dimension, and so forth. For a three-dimensional kd-tree the tree may be split, first in the x-dimension, then the y-dimension, then the z-dimension, and so forth. From an individual node in the kd-tree, the left branch contains values less than the value in that node (for the splitting dimension on that level), and the right branch contains values greater than the value in that node (for the splitting dimension on that level).
kd-tree 410 illustrates an example of a kd-tree generated from the coordinate pairs in set 405. In this case, the point (7, 2) is at a root node 412 of kd-tree 410. kd-tree 410 is split at the this level on the x dimension. Thus, the left branch from root node 412 includes nodes with an x value less than that of the root node 412. And the right branch from root node 412 includes nodes with an x value greater than that of the root node 412. The next level of kd-tree 410 includes nodes 413 and 414. The two nodes on this level are split in the y dimension. Thus, the tree to the right of the node 413 includes nodes with y values greater than that of the node 413 and the tree to the right of the node 413 includes nodes with y values greater than that of node 413. Similarly, the sub-tree to the left of node 414 (namely, node 416) includes nodes with y values greater than that of node 414. On the next level, splitting returns to the x dimensions, and a left branch and right branch may split the x dimension for these nodes. In the example kd-tree 410, however, no further nodes are present.
In one embodiment, a computational simulation of a conformation for a ligand and receptor stores the physical location (e.g., an (x, y, z) coordinate) for each atom in the ligand and each atom in the receptor as nodes in three-dimensional kd-trees. The resulting kd-tree structures may be used to:
As shown, the method 500 begins at step 505 where a first kd-tree is built to represent the physical positions of atoms in a receptor molecule. At step 510 a second kd-tree is generated to represent the physical positions of atoms in a ligand molecule. The relative positions of the receptor and ligand may be evaluated to determine whether any atom of one is too close to any atom in the other, for a given conformation. For example, the Lennard-Jones potential between the two atoms may be calculated. If the atoms are too close, then the conformation may be discarded from further evaluation.
At step 515, a spherical envelope may be generated. In one embodiment, the spherical envelope defines a region of space that encloses the atoms present in the ligand. As described in commonly assigned patent application “Envelope Technique for Exclusion of Atoms in an Hbond Check” (atty. docket ROC-9-2006-0210), the atoms in the ligand are usually distributed more compactly and may therefore be easier to define an envelope around than the atoms in a receptor, which is often much larger than the ligand. In various embodiments, the spherical envelope may be based on the center of mass of the ligand or on a geometric center of the ligand.
At step 520, the simulation may be configured to determine the position of the center of the spherical envelope and the radius of the spherical envelope. For example, the radius may equal the distance from the center of mass (or geometric center) of the ligand to the atom of the ligand furthest away from the center. Additionally, in one embodiment, the size of the envelope may be enlarged by an amount representing a distance over which hydrogen bonding interactions are expected to occur between the ligand and the receptor (typically a few angstroms).
At step 525, the simulation may search the first kd-tree (representing the positions of atoms in the receptor) to identify atoms of the receptor within the envelope. That is, the simulation may parse the first kd-tree to find all atoms of the receptor with a distance to the center of the spherical envelope smaller than the radius of the spherical envelope. Because of how the nodes of a kd-tree are constructed, however, not all of the atoms in the receptor need to be evaluated in these distance calculations. For example,
Returning to the method 500 of
In this simple example, the receptor and ligand are shown to include six atoms. Of course, in practice it is anticipated that the receptor will include substantially more atoms than the ligand. Nevertheless, this example illustrates that method 500 may substantially reduce the number of “too close” comparisons that are required for a given conformation of a ligand and receptor. Specifically, using a brute force approach to evaluate whether any of the atoms represented in this example were “too close” to one another would require 6×6=36 comparisons. And even limiting comparisons to atoms of the receptor within envelope 610 would still require 6×3=18 comparisons. However, because the kd-tree may be used to determine which atoms of the receptor are inside the spherical envelope and, for each such atom, the nearest neighboring atom of the ligand, the number of “too close” comparisons is reduced to the number of atoms identified at step 525. In this example, only three comparisons are needed. Thus, although some additional overhead is incurred in generating and parsing the kd-trees, it is more than offset by the reduction in the number of distance comparisons using a brute force approach.
While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.
This application is related to U.S. patent application Ser. No. ______, Attorney Docket No. ROC920060209US1, titled “Miss-Accumulation in a Binary Space Partitioning Tree”, filed May 1, 2007, by Gooding, et al; and U.S. patent application Ser. No. ______, Attorney Docket No. ROC92006021 OUS1, titled “Envelope Technique for Exclusion of Atoms in an Hbond Check,” filed May 1, 2007, by Mullins et al. These related patent applications are incorporated by reference herein in their entirety.
