New Algorithm Matches Any Tumor Cells To Best Possible Anticancer Treatments

Cancer patients don't have time to waste. Many go through several different treatments, however, to find one that is more effective against their particular type of tumor

Thus, an algorithm that could help rapidly sort molecular information about a patient's particular tumor and could help match this information to the right drug treatment would be a breakthrough of enormous value. Dan Theodorescu, M.D., Ph.D., a University of Virginia oncologist and cancer biologist, and Jae Lee, Ph.D., a computational biologist and bioinformatics statistician, have pioneered just such a system.

This work involved collaboration with colleagues at the National Cancer Institute, GeneLogic Inc. and the University of Virginia Computer Sciences Department. Using a panel of 60 diverse, human cancer cell lines from the National Cancer Institute (NCI-60), the researchers devised and tested an algorithm designed to match the best potential treatment(s) for a particular tumor in a particular patient.

Previously, the NCI-60 cell lines were used to screen more than 100,000 chemical compounds for their anticancer activity. These drug responses, however, were not definitely linked to clinical effectiveness in patients. Another issue is that the 60 cell lines did not include all important cancer types (for example, certain bladder cancers, lymphomas, and small cell lung cancers were not among the 60 lines studied).

The researchers investigated whether the drug sensitivity data of the 60 cancer cell lines could be extrapolated into useful information on other tumors or cancer cell lines. In fact, they found that their "coexpression extrapolation (COXEN) system" could be used to accurately predict drug sensitivity for bladder cancer cell lines to two common chemotherapies, cisplatin and paclitaxel.

"Even though this NCI cell set wasn't an exhaustive encyclopedia of cancer cells, we found we could use the available data to draw conclusions about other cell types we were exploring. The algorithm is a Rosetta stone for translating from the NCI-studied drugs to any other cell line or human tumor," says Dr. Theodorescu, director of the UVa Paul Mellon Prostate Cancer Institute and senior author of the study. "We believe we have found an effective way to personalize cancer therapy." The UVa research team was able to predict the clinical responses of breast cancer patients with treated with commonly used chemotherapies, docetaxel and tamoxifen.

The most exciting aspect of this research is that in addition to predicting patient responses to therapy, the COXEN algorithm can be used to discover effective compounds in any form of cancer. By the nature of the algorithm, which examines both cancer cells and drug activity at the molecular level, these newly discovered drugs should be effective in patients. This pre-screening for effectiveness using COXEN should greatly lower the failure rate of clinical trials testing new compounds. Likewise, as the drug discovery times are decreased in research laboratories, the cost of drugs also will come down. Basically it brings the chemists making the drugs much closer to the clinic, saving time.

Because the NCI-60 set of cells has been used to screen thousands of chemically defined compounds and natural extracts for anticancer activity, "we were able to make significant predictions about what compounds might work on real patients who might have other types of cancer," Theodorescu said. The researchers used the COXEN to screen 45,545 compounds, and they identified a several new compounds that have activity against human bladder cancer. To share this exciting capability with the scientific community, Dr. Lee is leading the development of a web-based COXEN system where investigators with genomic profiling data from cancer cells or patient tumors can obtain chemosensitivity prediction results on FDA-approved chemotherapeutic compounds.

Dr. Theodorescu is planning clinical trials for the new compounds against bladder cancer. Another planned clinical trial would examine patients with a variety of cancers receiving COXEN personalized, second-line drug combinations to beat their cancers, using FDA-approved agents. Many new and exciting discoveries remain to be made, even more quickly and at lower costs.

New Database To Help Develop AIDS Drugs

Researchers who are either developing drug treatments for AIDS or studying the virus that causes the disease have a new resource—an online database of AIDS-related protein structures just unveiled for public use by the National Institute of Standards and Technology (NIST).
Developed in collaboration with the National Cancer Institute, the HIV Structural Reference Databasewill receive, annotate, archive and distribute structural data for proteins involved in making HIV, the virus that causes AIDS, as well as molecules that inhibit these activities. Until now, much of this information was not widely available because it was unpublished. The new database contains data from both the published literature and from direct contributions by industrial and other laboratories.

The database will be especially useful in developing strategies for inhibiting the activities of the HIV protease (see image) that is essential for maturation of HIV. In addition, the database is expected to help scientists understand and circumvent the problem of mutations that make HIV resistant to certain drugs.

NIST scientists annotate the structural data with information from various sources and index—or classify—the entries so that users can reliably find particular structures. They helped to develop a novel technique for indexing HIV protease inhibitors, enabling scientists to rapidly and reliably get data on all enzyme-inhibitor complexes such as a mutant strain that is resistant to a particular drug.

NIST has a long history of producing, evaluating and disseminating chemical data and is increasingly applying this expertise in biosciences. The HIV database is a model for developing and testing new technology to annotate and standardize HIV inhibitor names, and for evaluating structural data for macromolecules.

New Database Screening Criteria Improves Identification Of Anticancer Drugs

Scientists in Indiana and Michigan have developed a better way of mining a vast computerized database for chemical nuggets that could become tomorrow's cance

The new "data mining" method pinpoints chemical structures with drug-like activity. It could speed the identification and development of new, more effective drugs against breast, prostate, lung and other cancers.

Computers have become a mainstay in the drug discovery process and have led to the identification of dozens of promising anticancer drugs. However, as the amount and complexity of information increases, new analysis methods need to keep pace.

In the new report, David J. Wild and colleagues analyzed data from the National Cancer Institute Developmental Therapeutics Program, a database of 40,000 compounds that have been tested against 60 tumor cell lines. The researchers identified a set of common structural features that can be used to more accurately predict which compounds are most active against cancer cells.

In a series of experiments, they showed that applying these new criteria significantly increased the accuracy rate of identifying drug-like molecules in comparison to standard screening methods.

Evolutionary Comparison Finds New Human Genes

Using supercomputers to compare portions of the human genome with those of other mammals, researchers at Cornell have discovered some 300 previously unidentified human genes, and found extensions of several hundred genes already known.
he discovery is based on the idea that as organisms evolve, sections of genetic code that do something useful for the organism change in different ways.

The research is reported by Adam Siepel, Cornell assistant professor of biological statistics and computational biology, Cornell postdoctoral researcher Brona Brejova and colleagues at several other institutions in the online version of the journal Genome Research, and it will appear in the December print edition.

The complete human genome was sequenced several years ago, but that simply means that the order of the 3 billion or so chemical units, called bases, that make up the genetic code is known. What remains is the identification of the exact location of all the short sections that code for proteins or perform regulatory or other functions.

More than 20,000 protein-coding genes have been identified, so the Cornell contribution, while significant, doesn't dramatically change the number of known genes. What's important, the researchers say, is that their discovery shows there still could be many more genes that have been missed using current biological methods. These methods are very effective at finding genes that are widely expressed but may miss those that are expressed only in certain tissues or at early stages of embryonic development, Siepel said.

"What's exciting is using evolution to identify these genes," Siepel said. "Evolution has been doing this experiment for millions of years. The computer is our microscope to observe the results."

Four different bases -- commonly referred to by the letters G, C, A and T -- make up DNA. Three bases in a row can code for an amino acid (the building blocks of proteins), and a string of these three-letter codes can be a gene, coding for a string of amino acids that a cell can make into a protein.

Siepel and colleagues set out to find genes that have been "conserved" -- that are fundamental to all life and that have stayed the same, or nearly so, over millions of years of evolution.

The researchers started with "alignments" discovered by other workers -- stretches up to several thousand bases long that are mostly alike across two or more species. Using large-scale computer clusters, including an 850-node cluster at the Cornell Center for Advanced Computing, the researchers ran three different algorithms, or computing designs -- one of which Siepel created -- to compare these alignments between human, mouse, rat and chicken in various combinations.

Over millions of years, individual bases can be swapped -- C to G, T to A, for example -- by damage or miscopying. Changes that alter the structure of a protein can kill the organism or send it down a dead-end evolutionary path. But conserved genes contain only minor changes that leave the protein able to do its job. The computer looked for regions with those sorts of changes by creating a mathematical model of how the gene might have changed, then looking for matches to this model.

After eliminating predictions that matched already known genes, the researchers tested the remainder in the laboratory, proving that many of the genes could in fact be found in samples of human tissue and could code for proteins. The researchers were sometimes able to identify the proteins by comparison with databases of known proteins. The discovered genes mainly have to do with motor activity, cell adhesion, connective tissue and central nervous system development, functions that might be expected to be common to many different creatures.