=Paper=
{{Paper
|id=None
|storemode=property
|title=Opportunities for Automated Workflow and Planning with Robot Scientists
|pdfUrl=https://ceur-ws.org/Vol-950/planlearn2012_abstract_Ross_King.pdf
|volume=Vol-950
}}
==Opportunities for Automated Workflow and Planning with Robot Scientists==
<pdf width="1500px">https://ceur-ws.org/Vol-950/planlearn2012_abstract_Ross_King.pdf</pdf>
<pre>
  Opportunities for Automated Workflow and Planning
                  with Robot Scientists

                                     Ross D. King

                  School of Computer Science, University of Manchester
                              Manchester, M13 9PL, UK
                           email: ross.king@manchester.ac.uk

     A natural extension of the trend of increased involvement in science is the concept
of a Robot Scientist [King et al. (2004),King et al. (2009)]. A Robot Scientist automat-
ically: originates hypotheses to explain observations, devises experiments to test these
hypotheses, physically runs the experiments using laboratory robotics, interprets the
results to change the probability of hypotheses, and then repeats the cycle (Figure 1).
Robot Scientists can autonomously execute high-throughput hypothesis led research.
In addition to automating experimentation Robot Scientists are well suited to record-
ing scientific knowledge: as the experiments are conceived and executed automatically
by computer, it is possible to completely capture and digitally curate all aspects of the
scientific process [King et al. (2009)]).
     The first Robot Scientists are Adam and Eve (Figure 2). The advances that distin-
guish Adam and Eve from other complex laboratory systems (such as high-throughput
drug-screening pipelines, and X-ray crystallography crystal-screening systems) are their
AI software, their many complex internal cycles, and their ability to execute individu-
ally planned cycles of experiments in high- throughput. Adam was designed to plan and
execute microbiological experiments. Adam was fully automated and there is no essen-
tial requirement for a technician, except to periodically add laboratory consumables and
remove waste. Adam was the first machine demonstrated to have autonomously discov-
ered novel scientific knowledge [King et al. (2009)]. Eve is designed to automate drug
discovery.
     There are many opportunities to improve Adam and Eve using automated workflows
and planning. Below I describe some of these opportunities in Adam. The target of




                          Fig. 1. Robot Scientist workflow, blue.
                                   Fig. 2. Picture of Eve.



Adams investigation were locally orphan enzymes. These enzymes catalyse biochemi-
cal reactions known to be in yeast, but for which the coding genes are unknown. Adam
used bioinformatic methods to abduce genes that could encode these orphan enzymes -
hypotheses. However, there are many orphan enzymes, and we manually (and greedily)
selected the enzymes for Eve to investigate that we estimated to be most experimen-
tally tractable based on availability of metabolites. This strategy was not necessarily
the best. It would be better automate and integrate such problem selection tasks using
planning.
    In developing Adam a large amount of effort was put into developing the software
to control the laboratory automation (robotics). But currently Adam dont use any work-
flows, with only some planning to fit investigations onto plates, and there remain many
opportunities to improve Eve using workflows and planning technology. A key practical
difficulty in applying workflows and planning is the lack of a good standard operating
systems for laboratory automation.
    Another area where there are many opportunities to improve Adam using automated
workflows and planning is in experiment interpretation. The function of most genes in
S. cerevisiae that when deleted results in qualitative effects (such as no growth) have al-
ready been discovered. The remaining genes result only in quantitatively effects when
deleted, e.g. they may have slower growth (bradytrophs), faster growth, higher/lower
biomass yield, etc. This required Eve to measure small quantitative difference, which
in turn required application of statistical experiment design and use of machine learn-
ing methods to determine which results were repeatable. To implement these methods
Adam used ad hoc scripts. Workflows and planning technology would have been better:
more efficient and enabling better annotation.
References
[King et al. (2004)] King, R.D., Whelan, K.E., Jones, F.M., Reiser, P.G.K., Bryant, C.H., Mug-
   gleton, S.H., Kell, D.B., and Oliver, S.G (2004) Functional genomic hypothesis generation
   and experimentation by a robot scientist. Nature, 427, 247-252.
[King et al. (2009)] King, R.D., Rowland, J., Oliver, S.G., Young, M., Aubrey, W., Byrne, E.,
   Liakata, M., Markham, M., Pir, P., Soldatova, L.N., Sparkes, A., Whelan, K.E., Clare, C.
   (2009) The Automation of Science. Science. 324, 85-89.

</pre>