How It Began — The DENDRAL Project

One of the first approaches for expert systems in general was developed by Stanford University at the request of the National Aeronautics and Space Administration (NASA) in 1965 [1]. At that time, NASA was planning to send an unmanned spacecraft to Mars that included a mass spectrometer for the chemical analysis of Martian soil. NASA requested software that would be able to automatically interpret the mass spectra to derive molecular structures. To solve this task, the Stanford team had to encode the expertise of a mass spectroscopist. 

This was the starting point for the DENDRAL project, initiated in 1965 by Edward Feigenbaum, Nobel Prize winner Joshua Lederberg, and Bruce Buchanan, in cooperation with chemist Carl Djerassi. The basic approach was to design software that mimics the concept of human reasoning and allows the formalization of scientific knowledge. 

DENDRAL consists basically of two subprograms Heuristic DENDRAL and Meta-DENDRAL. Heuristic DENDRAL uses the mass spectrum and other available experimental data to produce a set of potential structure candidates. Meta-DENDRAL is a knowledge acquisition system that receives the structure candidates and corresponding mass spectra to propose hypotheses for explaining the correlation between them. These hypotheses are validated in Heuristic DENDRAL. 

DENDRAL uses a set of rule-based methods to deduce the molecular structure of organic chemical compounds from chemical analysis and mass-spectrometry data. This process includes detecting structural fragments, generating structural formula from these fragments, and verifying it by determining the correlation between 

DENDRAL is written in list processing (LISP) and consists of a series of programs that shall be described here. 

---