In silico modeling is becoming a valuable and accurate prediction tool, despite its early, spotty reputation. The increase in information from metabolomics, proteomics, and genomics projects, plus clinical data, and better integration between bioinformatics and cheminformatics, are helping researchers build more complete and more complex models that are beginning to produce lab-proven results.
Several major companies using biosimulations are beginning to report their results in the literature, and a recent Kalorama Information (www.kaloramainformation. com) report, Informatics in Drug Discovery, reports that advances in computing power and new modeling techniques promise to shorten the discovery process by 15 percent. Return on investment figures remain elusive, and vendors content themselves with recounting anecdotal data.
Walt Woltosz, CEO, Simulations Plus (www.simulations-plus.com), tells the story of one generic drug companys bioequivalence trials for a complicated, controlled release dosage form. After failing two bioequivalence trials, the company developed a third formulation, and Simulations Plus ran the trial in silico. An intense, eight-week simulation study found the third formulation also would have failed.
The simulation also found that the first formulation probably would have passed if it had involved 100 subjects, Woltosz says. The 25 subjects involved werent sufficient for a good statistical sampling. The company saved a year or more of time and several million dollars by conducting an eight-week simulation, Woltosz says.
With stories like that circulating, in silico modeling shows tremendous potential for growth, according to the Kalorama report. It accounted for 7% of the informatics market share (three years ahead of predictions) in 2004, and Kalorama predicts an annual growth of 9% until at least 2008.
In silico modeling helps design better laboratory experiments and clinical trial protocols and define what should be measured, explains Mikhail Gishizky, Ph.D., CSO, Entelos (www.entelos.com).
It doesnt remove the need for experimental research and clinical trials. Once the question is well-focused, a model can be used to explore how heterogeneity in the patient population or changes in trial design will affect response. It gives researchers a faster way to run the what ifs.
Entelos is developing mechanistic mathematical models of human disease that have been used across the pipeline from early target identification through Phase IV clinical trial design. The firm is just beginning stage II of its collaboration with the American Diabetes Association to develop a model of type 1 diabetes based on the non-obese diabetic (NOD) mouse.
For this project, the key question is: why have therapies that have looked so promising in preclinical animal models failed in humans? Dr. Gishizky says. There are qualitative and quantitative differences between mouse and human physiology that significantly impact response to therapy. Very small differences mean a lot. By identifying and understanding those differences, researchers can better predict whether a therapy will be efficacious, and design more focused and effective drug trials.
In addition to the NOD mouse, Entelos has models for several human metabolic diseases (diabetes, obesity, and metabolic syndrome), inflammatory diseases (rheumatoid arthritis), and respiratory diseases (asthma and COPD).
Focusing on Metabolism
At Genomatica (www.genomatica.com), We work together with our partners to develop integrated computational and experimental research platforms designed around our advanced modeling and simulation technologies, says Christophe Schilling, Ph.D., president and CSO.
Genomatica focuses on metabolism because of its critical role in the majority of diseases. Genomaticas organism-specific cellular metabolism models are very good at predicting aspects of physiology, but they dont predict everything.
The company models certain types of metabolic problems, such as production and manipulation of small molecules by microbial cells and the effects of compounds on cellular metabolism under varying conditions.
The SimPheny (simulated phenotype) computational platform helps users create predictive metabolic models of organisms ranging from bacteria to humans by integrating organism-specific metabolic models with experimental data, and then simulate and analyze the metabolic capabilities within the context of the model.
It provides a comprehensive description of the metabolic process so researchers can manipulate genes or biochemical parameters, or alter the environment or their own hypotheses in silico.
Process optimization is one of the driving forces behind this technology, says Dr. Schilling, noting that it is used in industrial biotech to help develop new processes to produce small molecules and proteins, and to optimize existing microbial or enzymatic processes.
It may have particular value in such difficult products as amino acids, anti-microbial compounds, and in producing the base chemicals for polymers. So far, the company has developed about 12 models.
The next few years may see the development of a handful of models for microbial organisms and multicellular systems, followed eventually by models for human and other mammalian cells.
Compugen (www.cgen.com) is built around mathematical modeling, based upon the experimental and clinical data provided by its partners and its internal labs. That technology is helping companies make breakthroughs years ahead of the competition.
Novartis (www.novartis.com) is using Compugens models to study certain biological interaction networks. As Alon Amit, Ph.D., vp, science and technology, commercial operations group, elaborates, The goal is to demonstrate that we can incorporate data from diverse sources, analyze it, and predict new information regarding the relationship and timing events involved in specific types of biological networks.
If successful, the model will be able to predict relationships between proteins that would have been extremely difficult to discover experimentally on an individual basis. The integration of data from diverse sources is important because any single source, for instance, microarray data, often is very noisy and difficult to reproduce, Dr. Amit says.
Such modeling projects are lending validity to in silico modeling. There are specific diagnostic products in development based on highly reliable discoveries we made, says Dr. Amit. For instance, new splice variants of proteins for diagnostics were predicted by Compugen, which is working with several partners to further their development.
Currently, the company is developing qualitative models for specific pathways related to well-known drugs, with the hope of gaining insights into diversity in patient response to treatment. The pathways we work with are being intensely studied, Dr. Amit says. We created models based on published work and then improved their robustness and accuracy.
Simulations Plus just completed a major extension of its flagship product, GastroPlus. That product predicts dissolution/precipitation, transit, degradation, and absorption of drugs in the gastrointestinal tract, including the effects of transporter proteins like PepT1 and PGP, according to Woltosz.
Researchers using GastroPlus for physiologically-based pharmacokinetic (PBPK) simulations can get results for a 24-hour simulation in about 5 seconds, depending on the drug.
In PBPK simulations, each tissue is modeled independently to provide better animal-to-human scaling, better initial estimates of distribution of drug into various tissues, and total volume of distribution for new compounds, Woltosz says.
Simulation Plus Admet Predictor, updated in June with version 1.2.0, allows researchers to predict some 50 different properties from molecular structure. It is used for very high throughput in silico screening of large compound libraries and for estimates of key properties for single compounds. By using groups of artificial neural networks and averaging their outputs, it predicts properties critical to oral absorption as well as several pharmacokinetic properties and types of toxicity.
It has models for the maximum recommended therapeutic dose, estrogen receptors, toxicity in fathead minnows, tumor formation in rats, tumor formation in mice, and bacteria mutagenicity (the Ames test) and is working on four others that remain confidential.
Admet Modeler, also released in June, allows researchers to build structure-property models from their own data on their own servers and add them to Admet Predictor models, thus keeping the information in-house.
The technology combines automatic filtering and selection of descriptions, compound clustering for selection of training and test sets, and training a matrix of artificial neural net ensembles. The user selects the best ensemble for the final predictive model and sends it to Admet Predictor so it becomes a permanent new property prediction, enterprise-wide.
Simulations Plus released DDDPlus simulation software last spring for in vitro dose disintegration and dissolution studies for formulation scientists. Formulations for new active ingredients require only one calibration experiment before the software will predict how formulation changes affect the dissolution rate.
The application is similar to GastroPlus, but is only concerned with chemistry and physics, Woltosz says. The value occurs when you have such formulation changes as variations in amounts of active ingredients, excipients, and particle sizes. The program helps researchers simulate such changes and get results literally in seconds, he says.
MembranePlus, a similar product, is expected by years end to simulate and interpret Caco-2 and PAMPA in vitro membrane permeability. You can run exactly the same [physical] experiments in two different labs and get numbers that are 100 times different. With the simulation, you can understand the cause of these differences, Woltosz says.
Accelerys (www.accelerys.com) blurs the line between modeling and bioinformatics. It doesnt build models, per se, but builds the bioinformatic databases needed to make the models. It has virtual screening applications and products like Ligand Fit.
Using its bioinformatics capabilities, we can microscope down and look at how (components) interact at the atomistic level and thereby help researchers prioritize what they do in the lab, according to David Edwards, Ph.D., director of computational biology.
Biogen Idec (www.biogenidec.com) and Eli Lilly (www.lilly.com) proved the point, each taking a separate approach to the same challenge and detailing their results in a joint paper. Biogen Idec used Accelerys products to develop a pharmacophore in silico. Eli Lilly did the same thing, but in a wet lab. Both developed a very similar lead compound, but the in silico lead was identified in two months, while identifying the lead in the wet lab took 18 months.
Despite such success, Dr. Edwards says, We need to do a lot more mapping of the complexity in the human body. Mapping the human genome was a piece of cake compared to mapping the complexity and interactions of the human body.