ATPG for Cancer Therapy - Logic Synthesis for Genetic Diseases - page 84

Table 5.2 Drug selection for
multiple stuck-at faults

Net s-a Faulty PO Best PO Drug Vector Score

1,21 1,1 1111111 0000000 010001 84

4,9 1,1 1111111 0000000 000001 85

5,19 1,0 1111111 0000001 000110 74

6,8 1,1 1111111 0000000 000110 84

7,20 1,1 0000111 0000111 000000 56

8,21 1,0 0000111 0000000 000010 85

13,16 1,1 1111110 0000000 000100 85

1,3,6 1,0,1 1111111 0000000 000110 84

2,14,20 1,1,0 1111111 0000000 100001 84

4,7,17 1,1,1 1111111 0000111 010100 54

4,12,23 1,1,1 1111111 0000111 010000 55

8,9,11 1,1,1 0000111 0000111 000000 56

8,9,21 1,1,0 0000111 0000000 000010 85

12,18,20 0,0,0 0000110 0000000 000001 85

15,17,21 0,0,1 0000110 0000000 000001 85

5.4.2.3 Case 3: Fault Rectification with Minimal Drug Cost

When selecting drugs, there may be multiple drug combinations that may rectify a
fault, but where each drug has a different associated cost. We first assign weights to
drugs, according to their cost. For this case, we use the number of side-effects as the
drug's cost. Drugs AG825, lapatinib, Temsirolimus are assigned weights of 10, 15,
and 35, respectively, which correspond to their approximate number of side-effects
[ 17 ], [ 18 ]. However, drugs AG1024, U0126, and LY294002 have yet to under go
clinical trial and the number of side-effects is unknown. As such, these drugs are
assigned a weight 20, which is an average of the 3 previous weights.
In this GF example, Case 3 simulation provides the same results as in Case 2. This
is due to a lack of drugs that share paths in the circuit. In fact, for almost every non-
redundant fault, the best output state can only be achieved through one drug vector.

5.4.2.4 Case 4: Determining Therapy with Fewest Drugs and Best Coverage

Using the results from Case 2, we observe that the GF network has 13 testable
faults. For these 13 faults, we perform an All-SAT to find the top three scoring drug
combinations yielding the best output. All drug combinations are analyzed across all
single faults and presented in Table 5.3 showing drug vector, count of faults rectified,
and fault coverage. Drug vectors are ordered in increasing number of drugs selected.
From these results, we observe that with only 1 drug selected, the best coverage
is only 23 % of faults using lapatinib ( d 1 ) or Temsirolimus ( d 6 ). When allowing for
2 drugs, coverage increases to 77 % using the drug combination of U0126 ( d 4 ) and
LY294002 ( d 5 ). Finally, we achieve 100 % coverage of all testable faults when using
the 3 drug combination of U0126 ( d 4 ), LY294002 ( d 5 ), and Temsirolimus ( d 6 ). When
the single stuck-at fault location is unknown, these selected drug combinations will
be the most effective for therapy and for preventing the proliferation of cancer.