BioInformatics Reviewed Links
[PAGE 1 PAGE 2 PAGE3 PAGE4 PAGE5]
Link: http://www.nci.nih.gov/bip/spieppr.htm
Reviewed by: Mohammed Kuddus
ЖЖЖЖЖЖЖ NCI Initiatives in Computer Aided Diagnosis
This paper reports the progress of an image database for lung cancer
screening that uses spiral X-ray CT being investigated by the National
Cancer Institute. The best method of use involves the cooperation of
investigators willing to join a collective of institutions to build a
database for public use and research, such as the National Library of
Medicine Visible Human Project, initiated in 1989, and the Human Brain
Project, begun in 1993. The eventual goal is to build norms for
generating the database and allow it to be used for assessing computer aided
diagnosis software methods. Spiral CT of the lung was the initial focus of
interest to aid early intervention because the imaging database was used for lung
cancer screening for patients at high risk. The more efficient the screening
becomes, the more potential there is in using Computer Aided Diagnosis
methods for large-scale cancer screenings. Lung imaging offers an ideal
physical model for 3D Computer Aided Diagnosis methods for both
detection and classification and any change in the CT images over time will
provide improved early cancer detection or better categorization.
Link: http://www.techreview.com/magazine/oct01/cohen_pro.asp
Reviewed by: Christopher Conway
ЖЖЖЖЖЖЖ The Proteomics Payoff
This article is an overview of the field of proteomics: the history,
the potential, the hype and the reality. The discussion centers on the
possibility for a "human proteome project" to follow the mapping of
the genome and proteomics' applications in drug discovery and medical
treatments.
The word "proteome" was coined in 1994 to mean "all the proteins
expressed by a genome, cell or tissue." Since the genome is essentially a
blueprint for creating proteins, a fully sequenced human genome leads inevitably
to the question of the proteins coded for within. The p! roblems of gene
expression, protein formation and structural complexity make "mapping
theproteome" a much trickier proposition than simply sequencing base pairs
of DNA. DNA is essentially a linear combination of just four nucleotides
found in the same form in the same place in every cell of an organism.
Proteins, in comparison, are expressed selectively in different tissues. They are
formed from combinations of 20 different amino acids and every protein
has a unique three-dimensional structure. A human proteome map would need
to identify, sequence and structurally analyze every protein expressed in
every tissue of the human body.
The technology available for protein analysis will not allow f! or an
undertaking of that scale. Very few automated tools exists for working
with proteins. Researchers use gel electrophoresis to separate them, mass
spectrometry to analyze their components and x-ray crystallography to
determine their three-dimensional structure. Each of these techniques
is time-consuming and each of them has its flaws. For instance, x-ray
crystallography relies on the ability to crystallize a protein --
something which is often difficult, or perhaps impossible, to do. There exists no
tool for working with proteins that is as efficient as the automated
DNA sequencer.
Given these limitations, researchers are concentrating on relatively
small slices of the problem. ! The Alliance for Cellular Signaling is working
to identify the proteins that are involved in the cellular signaling of
two types of mouse cells within ten years. Large Scale Biology is working
on a Human Protein Index that will identify significant proteins in
"medically relevant" tissues. Celera is concentrating on identifying proteins
associated with specific diseases.
Link: http://www.wisdom.weizmann.ac.il/~udi/DNA5/turing5.html
Reviewed by: Christopher Conway
ЖЖЖЖЖЖЖ Blueprint for a Biomolecular Computer
The paper describes a mechanical device conceived by Ehud Shapiro which
models Turing machine, i.e., a universal programmable computer. It may
be possible to implement the device as a biomolecular machine that
might operate in vivo to construct synthetic polymers based on input from its
cellular environment.
A Turing machine is a conceptual device which consists of a read/write
head positioned on a linear tape marked with symbols and a set of state
transition rules of the form (Current State, Current Symbol, New State,New Symbol, Move Left/Right). In other words, a Turing machine readingan input symbol from a tape can move to a new state, output a new symbolto the tape and move the read head left or right. Church's thesisconjectures that any function which is computable is computable by a Turingmachine.Mr. Shapiro's machine uses a polymer of "alphabet molecules" as itstape. Each alphabet molecule consists of a side-group representing a symboland left and right links to form the polymer tape. State transitions arerepresented by "transition molecules" consisting of recognition sitesfor the previous symbol (alphabet molecule) and state (transitionmolecule), a side-group representi! ng the new state, a location site for the newsymbol (alphabet molecule) and left and right links that allow it to beembedded within the polymer tape.A program for the machine consists of a set of "loaded" transitionmolecules (i.e., transition molecules with attached alphabet molecules,representing codified state transitions). At any given step in thecomputation, the polymer tape will have a transition molecule embeddedat some point, with adjacent alphabet molecules to the left and to theright allowing for either a left or right state transition (i.e., atransition of the form "S0,a->b,S1" (left) or "a,S0->S1,b" (right)). If there isat least one transition! molecule that accepts either of those states, itwill bond from above, displacing the previous transition molecule and thematched alphabet molecule and incorporating itself into the tapepolymer. (See figures: http://www.wisdom.weizmann.ac.il/~udi/DNA5/turing5_figures/index.htm)As the execution states "stack up" a trace polymer is created containingevery previous state and "erased" symbol. This is not necessary for aTuring machine, but it makes the machine's computations reversible, andallows for program analysis and error detection.
Link: http://www.arstechnica.com/reviews/2q00/dna/dna-1.htmlReviewed by: Christopher ConwayЖЖЖЖЖЖЖ DNA Computing: A PrimerThe article is a detailed explanation of how Leonard M. Adleman usedDNA to solve the traveling salesman problem, a popular restatement of thedirected Hamiltonian path problem. The problem posits a salesman whomust plan a trip starting in one city, ending in another and visiting once-- and only once -- n cities in between. It is a classic example of acomputationally intractable problem, as its complexity growsexponentially with the number of cities. Adleman's strategy was to use biochemicalmethods on DNA to massively parallelize the computation.!Adleman encoded the cities of the problem as unique strands of DNA 20base pairs long. He then encoded the paths between cities as the complementof the last half of the strand representing the first city and the firsthalf of the strand representing the second. So, for instance, if New Yorkwas encoded as a 6 base pair long strand 'ATGCCG' and Los Angeles wasencoded as 'GCTACG', the (directed) path between them would be the complementof 'CCG' and 'GCT' taken together, or 'GGCCGA'. When these strands of DNAhybridize the strand representing the path will glue together thestrand representing the cities -- DNA hybridizes with its complement. Thismeans that if you put the strands representing all of the cities and thestrands representing! all of the paths into a test tube and allow them tohybridize,the path strands will connect city strands and you will end up withmany longer strands of DNA representing potential solutions to the problem.this is equivalent to performing an exhaustive search for everypossible answer to the problem. But whereas on a traditional serial processorsuch a search would be computationally infeasible, DNA by its natureperforms millions of these "searching operations" simultaneously.Next, Adleman used PCR to selectively amplify only strands that beginand end with the correct city encoding. Then he used gel electrophoresis toselect only strands that were exactly the right length. Finally, he putthe strands through several stages of affinity purification to
eliminate any strands that did not contain every city. The solution can then be
read using either a DNA sequencer or graduated PCR to determine the order in
which city encodings occur in the solution.
Adleman used this technique to solve the traveling salesman problem for
seven cities and it took him sevenЖ days to determine the answer. This
is not competitive with traditional modern microcomputers. In addition, as
the number of cities grows the technique begins to require impractical
quantities of DNA. However the experiment was quite effective in
demonstrating that, for certain types of problems, DNA can be
effectively used as a computing medium with the potential for massively parallel
computation.
Link: http://pubs.acs.org/hotartcl/mdd/00/may/razvi.html
Reviewed by: Lu Tang
ЖЖЖЖЖЖЖ High-throughput Genomics
This article discusses how and why single nucleotide
polymorphisms(SNP) genotyping affects the
pharmaceutical industry. Genomics is fundamental to
the success of the biotechnology industry. SNPs can be
used to enhance drug discovery and development.
Seizing a prominent role in the biopharmaceutical
community, SNPs are binary elements of genetic
variability in the human genome and function as
signatures for different biological traits.
SNPs associated with a single disease offer clues
about the underlying pathology of the disease state,
within what was once considered a single disease. A
major focus of the SNP effort is on using small
variations in gene sequences as makers for defining
populations exhibiting a given phenotype.
Pharmacogenomics is the application of this genetic
knowledge by targeting therapies on the basis of
genomic compositionthe expression of a given
haplotype. The major drivers for the SNP effort in
health care are (1)defining genetic regions and
targets for therapeutics; (2)stratifying patient
populations according to expression of SNP markers
(and corresponding biological
phenotype);(3)positioning drugs into the appropriate
subsectors of patien! t populations.
In the near term, SNPs could potentially help optimize
clinical trials through patient stratification. In
addition to the near-term opportunity for SNP
genotyping, it can be deployed in other market
segments, including pharmacogenomics (which stratifies
patients according to their responder/nonresponder
status to a given drug), molecular diagnostics, and
predictive medicine. We estimate that these segments
represent the long-term upside in the SNP genotyping
marketplace.
Instrumentation for SNP discovery, characterization,
and genotyping will be deployed throughout the
low-thoughput, sporadic use, to high-throughput,
production-scale instrumentation. Informatics tools
and algorithms enable data manipulation and mining.
Given the amount of data to be manipulated, robust
information technology tools are needed. Wet chemistry
must keep pace with the information deluge that is
likely to result as the private and public genome
projects develop more content. Genotyping is the key
process by which SNPs are harvested into commercially
useful information in the form of biological
associations.
DNA arrays are glass slides or membrane filters
containing many immobilized ! DNA samples. DNA
microarrays are best known as tools for monitoring
gene expression. The advantage of microarray
technology is that production is relatively well
defined and automated, which has made it appealing to
those working in the biopharmaceutical community.
Mass spectrometry is an accurate and potentially
high-throughput approach to genotyping. It relies on
minute differences in molecular mass to identify DNA
fragments. It maybe possible to bring down individual
genotype costs in the future, and the approach is
flexible and may be automatable for production-scale
genotyping. But it is a complicated system.
!
High-efficiency fluorescence polarization(HEFP) is a
homogeneous mix-and-read assay process without any
washing or separation steps. HEFP is based on
discriminating the rapid molecular rotation of small
molecules from the slower rotation of larger species.
High-throughput screening of SNPs is an important new
field. SNP genotyping technologies are maturing, and
companies are commercializing them in the form of
products. These successful technologies offer proven
and open solutions, deliver robust results, are cost
effective and easy to implement, and yield high value.
Link: http://www.geml.org/
Reviewed by: Christoforos ChristoforouЖЖЖЖЖЖЖЖЖЖЖЖЖ
Introduction to Gene Expression Markup Language.
This paper introduces Gene Expression Markup Language
(GEML).
GEML is an Extensible Mark Language (XML)-based set of
tags and aims to provide a standard method of
exchanging gene expression data along with associated
gene and experiment annotations. Files with GEML
format can store data about patterns, profiles, andhybridization information for gene expressionanalysis, providing a standard method for collectingand transferring DNA microarray and gene expressiondata.GEML provides standardization for storing andtransferring between different systems and databases.It also keeps track of the data collection methodologythat was used, enabling normalization, integration andcomparison of data across methodologies. Furthermore,it is independent of any particular database schema.Given the large amount of data derived from geneexpression studies, the need for this data to bestored, transferred, and anal! yzed, has increased.However, up until now there has been no standardmethod of storing and exchanging this data. As furtherresearch is conducted in the area of gene expression,it becomes more and more important for the fields ofpharmaceuticals and the biotechnology field, and also,for academic researchers to be able to share datacooperatively. The need for a standard method ofexchanging the data and the associated annotation hasbecome paramount. GEML can contribute to the needs anddemands of the field of gene research.
Link: http://dodo.cpmc.columbia.edu/predictprotein/Reviewed by: Mohammed KuddusЖЖЖЖЖЖЖ Predicting secondary structure from protein sequencesThe web page presents predicting secondary structure from proteinsequences by PHD(Profile network from HeiDelberg) method. Protein structure playsan important role for understanding and use of sequence data. A knowledgeof the protein structure behind the sequences often makes clear whatmutational constraints are imposed on each position in the sequence and cantherefore aid in the multiple alignment of sequences and the interpretation ofsequence patterns. Protein structure is intrinsically hierarchic in itsinternal organization. The highest level in this hierarchy isconstituted by complete proteins or assemblies of su! ch proteins, which becomesubdivided through domains via super-secondary structure to secondary structure atthe lowest hierarchical level. At the level of protein secondary structure,the elements are not only crucially dependent on their amino acidcompositions, but unlike domain and higher-order structures, are also very muchcontext dependent; i.e. they rely critically on the substructures in theirenvironment. It is because of this context dependency, that predictingprotein secondary structure is a difficult task, which after threedecades of research has not attained the accuracy on which tertiary structurecan be based. However, that some successful prediction of higher-orderstructure, based on a knowledge of the secondary structure, have been achieved.Secondary structure prediction has generally been formulated for threestates, helix, strand and coil. This holds also for recent versions ofthe early and popular GOR method, which considers the influence andstatistics of flanking residues on the conformational state of a selected aminoacid to be predicted. In 1987, Zvelebil et al. for the first time exploitedmultiple alignments to predict secondary structure automatically by extendingthe GOR method. As s result, the current state-of-the-art methods all use inputinformation from multiple sequence alignments. Neural network areorganized as interconnected layers of input and output units, and can alsocontain intermediate (or 'hidden') unit layer! s. Each unit in a layer receivesinformation from one or more other connected units and determines itsoutput signal based on the weight of the input signals. The PHD method(Profile network from HeiDelberg) combines the added information from multiplesequence information with the optimization strength of the neuralnetwork formalism. The method makes use of three consecutive complete neuralnetworks:(a) The first network produces the first raw 3-stateprediction for each alignment position. The output of the first network for eachalignment position is three probabilities for three the states (helix,strand, and coil).(b) A second netwo! rk refines the raw predictions ofthe first level by filtering the 3-state probabilities for each alignmentposition based on the probabilities of the flanking positions. Theoutput of the second network comprises for each alignment position the threeadjusted state probabilities.(c) The first two networks perform the basic predictionof the secondary structure associated with a query multiple alignment.However, as the networks can be trained in various ways, PHD employs anumber of separately trained consecutive network pairs ((a) and (b))and feeds their predictions (3-state probabilities) into a third networkfor a so-called jury decision.The predictions obtained by the jury undergo a final filtering toDelete predicting helices of one or two residues and changing those into coil.The method was trained on a non-redundant set of 130 alignments from theHSSP database, each containing one sequence with a known structure. If thePHD web server is given a single sequence for prediction, it performs aBLAST-search to find a set of homologous sequences and aligns thoseusing the MAXHOM alignment program. The resulting alignment is then fed intothe actual PHD neural net algorithm.The accuracy of computerized prediction methods can be enhanced furtherif such reasoning with higher order structure is formalized andincorporat! ed in the prediction mechanisms. If some easy benefits comes from thesteadily increasing structural protein data that can be used to better train andtune the statistical methods. The current availability of the predictionmethods optimizes the chance for development of sensitive consensus methods.
Link: http://www.marmulla.com/compengl.htmReviewed by: Xiaodong (Sheldon) ZouЖЖЖЖЖЖЖЖЖЖЖЖЖcomputer-assisted surgeryЖЖЖЖDue to the expanding application of computertechnology in medicine new methods are evolving formedical diagnosis, education and training, as wellas of surgical treatment planning, assistance andassessment. The planning of complex surgicalprocedures necessitates highly reliable computerizedmodels of the human anatomy. Repositioning osteotomyis! a frequently used method correcting malpositionsin orthopedic surgery and traumatology. Computertomography, stereolithography models and tele-X-raysare used in planning. However, the precisionachieved in the planning phase is usually nottranslated to patients.Ж The segment navigator SSN is a navigation systemwhich allows for the computer-assisted correction ofmalpositions. It consists of an infrared-positioningdevice, two dynamic reference frames DRF, aninfrared-pointer and an infrared-camera. All dataare displayed numerically and graphically on themonitor of the SSN workstation.ЖЖA Laboratory Unit for Computer Assist! ed SurgeryLUCAS is used for planning surgery in thelaboratory. LUCAS requires only a scout view CT. Apreparatory operation to implant bone markers visiblein X-rays and a further planning CT scan showing thebone markers - which were necessary in previoussystems - are not required for the LUCAS- andSSN-system. This reduces significantly the radiationexposure of the patient and the costs of surgicalplanning. Even the measurements of anatomicallandmarks in the surgical site which are timeconsuming and reduce the accuracy are not required forthe SSN-system since the position of theinfrared-transmitters is already known during surgicalplanning on the LUCAS-workstation. This makes ! thesurgical approach faster and much more precisely. Thesurgical planning data are transferred to the surgicalsite using a data file and an individual surfacepattern which fits to the surface of the navigatedbone segment:Ж The data file is exported from the LUCAS-workstationto the SSN-workstation. The plannedspatial displacement of the infrared-transmitters issaved in this file.Ж The individual surface pattern carries theinfrared-transmitters. This pattern is themechanical interface between infrared-transmitters andnavigated bone segment.Ж The individual surface pattern can be polymerizeddirectly on a small stereolithographic model of thenavigated bone segment. The surface pattern can aswell be generated as negative form from a CT dataset using a CAD-CAM-system. To summarize, LUCAS andSSN allow for the computer-assisted correction ofmalpositions, positioning of artificial joints andimplants. In principle, the systems can be used in allfields of surgery.Ж The SSN is based on an infrared positioningdevice such as the Surgical Tool Navigator (STN) andthe Surgical Microscope Navigator (SMN) manufacturedbyCarl Zeiss.Ж The infrared positioning device is connected toa Microsoft Windows NT 4.0 Workstation on a HewlettPackard NetServer LD Pro.Ж The software of the Surgical Segment Navigator(SSN)and the Laboratory Unit for Computer AssistedSurgery (LUCAS) is written by R|diger Marmulla andcompiled with Microsoft Visual Studio 6.Ж The method of computer assisted bone segmentnavigation is described with interactive software,multimedia files, movies and source code on a CD.
Link: http://www-2.cs.cmu.edu/People/tissue/layering.htmlReviewed by: Changyou YuЖЖЖЖЖЖЖЖЖЖЖЖЖЖ Tissue Engineered BoneThis web page discusses a new Solid Freeform Fabrication assemblymethod that is crucial to create an advanced CAD/CAM bioreactor system capableof growing large-scale,customized bone substitutes.This method isdeveloped by an interdisciplinary team from The Robotics Institute and The Institutefor Complex Engineered System of Carnegie Mellon University and theUniversity of Pittsburgh Medical Center.To synthesize bone tissue, a CAD model of the desired bone substitutewould first be derived from CAT or MRI data of the patient. The syntheticbone would then be fabricated, in-vitro, in an advanced CAM bioreactor! bydepositing layers of biodegradable scaffolding material whilesimultaneously embedding donor cells and growth factors within the layers. Syntheticvasculature would also be embedded within the scaffold as it is beingbuilt up. Biological and mechanical stimulus, would be provided to controlthe growth of the tissue until it was mature enough to be removed from thebioreactor and implanted into the patient.The traditional scaffold-guided tissue regeneration method is oftenlimited in practical thickness partly due to the difficulty in getting cellsdeep into interior regions of scaffolds. In the new method developed by thisgroup, thin(! approx.1 mm thick)and prefabricated cross sections canfirst be seeded with cells and/or growth factors, then these sections arestacked up to form 3-D structures by mating them together with biodegradable ornon-biodegradable fasteners. Then normal tissue growth across thelayers, in vivo, fuses the assembly together as the scaffold degrades.In one of the experiments conducted by the group,two 3D constructs werecreated, one with 5 seeded layers, the other with 5 unseeded layers.They were both implanted into the rectus abdominis muscle of the male NewZealand White rabbits. Then the specimens remained implanted for eight weeks.The result revealed that ther! e was a statistically significant greateramount of bone formation in the implants seeded with cells than those unseededimplants.This assembly approach also provides a way to integrate differentmaterials with different microstructures, it allows prefabricated vascularconstructs to be embedded and assembled into the scaffold as it is being built upas one source for capillary sprouting. As a result,more complex structuressuch as complete bone, tendon, muscle constructs could be obtained by thismethod.
performed directly between the bone and the robot,
which in turn guides the surgical tools. Then they use
the hierarchical multi-resolution approach and a level
of detail data model to speed up the computation. To
ensure the required accuracy of 1 deg ratation and 1mm
translation, they explore the subject of optimal
number of sampling points and their location. The
search for the best sampling points is viewed from the
grasping theory point of view. The contacts between
the fingertips and the grasped object are modeled as
frictionless point contacts. In order to determine the
optimal set of sampling points for each set of points,
they look for the bone motion which is minimally
detected by the sensor. A! mong these motions, they
choose the set of contact points that are maximally
dislocated by the minimal motion. This configuration
of the grasp gives the best stiffness properties;
hence, the points of contact are the best candidates
for sampling during registration.
Link: http://robotics.eecs.berkeley.edu/~mdownes//surgery/surgsim.html
Reviewed by: Ying Liu
Virtual Environments for Surgical Training and Augmentation(VESTA)
As computer graphics and processor technologies
advance quickly, researchers are trying to increase
high-fidelity computer simulations of surgical
procedures. They developed laparoscopy's surgical
technique that has greatly reduced the time necessary
for the patient to heal after surgery. A laparoscopic
simulation device is an essential component of
surgical simulation that is to model the deforma! tion
of soft tissue accurately and in real time. You can
move each tool further into or out of its virtual
hole, or twist each tool around its longitudinal axis.
Laproscopic camera training simulation is in minimally
invasive surgery of the abdomen, surgeons watch a
video image from a rigid lens endoscope inserted
through an incision in the skin. Such an endoscope
uses a camera whose lens is aligned with the body of
the scope, and surgeons are fairly well experienced
with using the device. Laparoscopic cholecytectomy
simulation is a laparoscopic surgical procedure to
remove a patient's gallbladder. It provides a virtual
environment for training surgeons for this new
surgical technique. Th! ey have created realistic
textures for the anatomy in their environment by
capturing images from videos of actual surgical
procedures. The physics model in their system is
modeled as spring-damper matrices in which nodes in
the matrix are masses.
The researchers achieve the proper behavior for
cutting and stapling ducts by altering the mesh for a
duct when it is cut or stapled. They have completed to
design a system suitable for training surgeons to use
the angled lens endoscope. Thus, a physical
environment can beset up to train the surgeons and a
similar environment can be mimicked using their
virtual environment system. They use the physics-b! ased
models to help of local integration, which is shown to
be a successful way to significantly reduce the time
complexity. But the researchers think that it exists
some problems. The laparoscopic immersion device that
they have used as their interface does not yet support
force-feedback. Without force-feedback it is very
difficult to keep the position of the virtual
instrument calibrated with that of the real device
when the device can move through its entire range of
motion unimpeded. They hope to eventually get a
laparoscopic device that includes force-feedback and
integrate it with our system or experiment with using
Phantoms as input devices.