How AI is helping researchers develop antibiotics to fight drug-resistant infections — Transcript

Full transcript

1. 0:00Drug resistant infections are a major public health threat around
2. 0:03the world. Responsible for more than a million deaths each year.
3. 0:07So scientists are constantly trying to find and develop new antibiotics.
4. 0:12Miles O'Brien reports on how researchers now say AI is helping to
5. 0:17speed up their search.
6. 0:21This is the front line in a biological arms race to salvage the crumbling
7. 0:27foundation of modern medicine, antibiotics. They make surgery routine,
8. 0:32protect cancer patients and turn once deadly infections into minor
9. 0:37inconveniences. Let's take a look at gonorrhea first. The discovery
10. 0:41of penicillin changed everything, not least the treatment of sexually
11. 0:46transmitted disease. It is a great boon to the private position.
12. 0:50to clinics and of course with the patient. But success comes with
13. 0:54a fatal paradox. The more we deploy this life saving medicine, the
14. 0:59less effective they are in the long term. Meliss Anaar is a clinical
15. 1:04microbiologist at Massachusetts General Hospital. It's unlike any
16. 1:08other drug where when we use antibiotics, we by definition lose them
17. 1:12because we're in this constant race with bacteria, where the bacteria
18. 1:16can evolve resistance to our antibiotics in real time. Bacteria develop
19. 1:21resistance to antibiotics through a simple process of evolution.
20. 1:27In any infection, there are millions of bacteria, and some have mutations
21. 1:31that help them survive a drug. When antibiotics are used, they kill
22. 1:36the vulnerable bacteria, but the resistant ones survive, multiply,
23. 1:41and spread. Over time, these resistant strains become dominant, making
24. 1:46the drugs less effective or even useless. There's kind of this never
25. 1:51ending war, these microorganisms they're just not going to quit.
26. 1:54They do not quit, but neither does she or her colleagues here at
27. 1:58the Broad Institute of MIT and Harvard. Historically, to find and
28. 2:03test new antibiotics, researchers would gather up some molecules
29. 2:06stored in a deep frozen library of compounds and then apply them
30. 2:11one by one to a pathogen to see what can either stunt its growth
31. 2:14or kill it outright. biomedical engineer Jim Collins runs the lab.
32. 2:20It's a laborious process. It's very much kind of searching for a
33. 2:24needle in a haystack that's an expensive haystack. They would find
34. 2:28a promising molecule less than 1% of the time.
35. 2:32Enter artificial intelligence. Collins and his team trained a deep
36. 2:36neural network to analyze the chemical structures of molecules, and
37. 2:41bond by bond, substructure by substructure, so kind of these ba sticks
38. 2:45that we all remember from our high school chemistry days could associate
39. 2:49those properties with being antibacterial or not antibacterial. And
40. 2:52now a model was trained that you could feed a new compound structure,
41. 2:56these balls and sticks, and the model could calculate, could this
42. 2:59make for a good antibiotic? They applied the AI to the library of
43. 3:036000 compounds here at Broad to find molecules that would make for
44. 3:07antibiotics that are effective, are not toxic to human cells and
45. 3:12have not yet been discovered. Of the 6000, only one molecule. satisfied
46. 3:17all three criteria, and it's a molecule we call halison,
47. 3:21and Hallison turns out to be a remarkably potent new antibiotic.
48. 3:26that kills multidrug resistant, extensively drug resistant, and pan
49. 3:30resistant bacteria through a new mechanism of action. Then they deployed
50. 3:34AI and computing power to virtually generate and screen 70 billion
51. 3:39theoretical molecules to test how they might behave. In this case,
52. 3:44AI is doing something computational chemist Andreas Lutins does intuitively.
53. 3:50Instantly seeing what molecules might work against a pathogen based
54. 3:55on how the balls and sticks line up. I usually work with the bucket
55. 3:59system, and then order scoring system like there's stuff that I,
56. 4:03I really like, stuff that I hate. No, no, no, no, no,
57. 4:10maybe. No, this is like small molecule tinder swipe right, exactly,
58. 4:16yeah, it is a molecular dating app. You're trying to find compounds
59. 4:21that you like and you can go quite fast in this selection process.
60. 4:25He may be fast, but he can't match the scale and persistence of a
61. 4:30machine. Depending on how much coffee I get and how long I can stay
62. 4:34awake. It will beat me eventually, which brings us back to Meliss
63. 4:38Annaar and the pathogens she is focused on Nigeria gonorrhea. That's
64. 4:44gonorrhea right there, huh? This is gonorrhea. So you see these looks
65. 4:48nasty colonies, grayish, untreated, the sexually transmitted disease
66. 4:53can escalate into serious, sometimes irreversible health problems.
67. 4:57The bacterium is resistant to nearly everything outsmarting new drugs
68. 5:02about every 5 years or so. The currently prescribed antibiotic, ceftriaxone,
69. 5:08is nearing the end of its efficacy. Finding something new and potent
70. 5:13is an urgent problem
71. 5:15The AI system screened 45 million chemical fragments from that vast
72. 5:20universe, the most promising chemical seed was used to generate an
73. 5:25additional 7 million candidates. After rigorous filtering, two of
74. 5:30these new compounds were synthesized and tested against real bacteria
75. 5:35in the lab. Is it really killing the bug in vitro, and is it not
76. 5:39harming human cells? Pink means there's bacterial growth, and blue
77. 5:43means that the growth is inhibited, so we want to see lots of blue.
78. 5:47This looks like a home run here, so this one looks good. This one
79. 5:50was not as successful. In the end, there was one novel compound that
80. 5:55killed drug resistant gonorrhea without causing serious harm to human
81. 6:00cells, not only can we find these antibacterial compounds, but they're
82. 6:04actually inhibiting new targets. This was created from scratch based
83. 6:09on what it learned from. existing small molecules and drugs. Globally,
84. 6:15drug resistant infections kill more than 1 million people each year,
85. 6:20and if nothing changes, the experts predict that number will increase
86. 6:24by 50% by 2050. So is resistance right now moving faster than the
87. 6:30research to try to address it. Resistance had been developing faster
88. 6:35than the research and development. that had been underway, but I
89. 6:39believe that this infusion of AI has now changed the game. We now
90. 6:42have tools that have dramatically expanded our ability to both to
91. 6:46discover and design new antibiotics. None of this will increase the
92. 6:49speed of clinical trials in human beings, nor should it, and it does
93. 6:54nothing to incentivize big pharma to manufacture new antibiotics
94. 6:59which don't generate big profits, but artificial intelligence might
95. 7:04be one way to begin recharging a crucial pipeline that has dried
96. 7:09up. For the PBS Newshour, I'm Miles O'Brien in Cambridge, Massachusetts.