Bris­tol My­ers Squibb is turn­ing to one of the star up­starts in the ma­chine learn­ing world to go back to the draw­ing board and come up with the dis­ease mod­els need­ed to find drugs that can work against two of the tough­est tar­gets in the neu­ro world.

Daphne Koller’s well-fund­ed in­sitro is get­ting $70 mil­lion in cash and near-term mile­stones to use their ma­chine learn­ing plat­form to cre­ate in­duced pluripo­tent stem cell-de­rived dis­ease mod­els for ALS and fron­totem­po­ral de­men­tia.

Then they’ll use those in­sights to start build­ing new drugs for those two ail­ments; a com­plex, ground-up ap­proach that has al­ready won a close al­liance with Gilead.

Suc­cess would trig­ger up to $2 bil­lion in mile­stones, run­ning a gamut of re­search and com­mer­cial goals.

Richard Har­g­reaves

“We be­lieve that ma­chine learn­ing and da­ta gen­er­at­ed by nov­el ex­per­i­men­tal plat­forms of­fer the op­por­tu­ni­ty to re­think how we dis­cov­er and de­sign nov­el med­i­cines,” said Richard Har­g­reaves, the chief of the neu­ro group at Bris­tol My­ers, who made the leap from Cel­gene.

Koller’s been mak­ing great strides with a new tech­nol­o­gy that has gained im­mense in­ter­est, but still has a long way to go to prove it­self as the ma­jor play­ers start fig­ur­ing out how to in­te­grate ar­ti­fi­cial in­tel­li­gence and ML in­to their game plans for drug dis­cov­ery and de­vel­op­ment.

As she told me in in­ter­views for the End­points 11 awards, the field has been in­tense­ly hyped by a slew of new play­ers. From her per­spec­tive, the tech holds lots of promise, but there’s a tremen­dous amount of ground­work that has to be done to glean the nec­es­sary da­ta for it to work. And it starts with the right dis­ease mod­els.

I’ve looked for in­stance at cel­lu­lar phe­no­types and I’ve built a mod­el that aligns those with hu­man clin­i­cal out­come: Can I take a group of ge­net­ic back­grounds that I’ve nev­er seen be­fore and pre­dict clin­i­cal out­comes for those pa­tients? That is a con­fi­dence build­ing mea­sure. It says I built a dis­ease mod­el that is ac­tu­al­ly pre­dic­tive of what we see in hu­mans. By the way, an­i­mal mod­els, they don’t even get asked that ques­tion in a lot of cas­es, which I find rather shock­ing.

On top of that, the ex-Stan­ford pro­fes­sor has been frank about un­der­stand­ing the tech bet­ter than drug de­vel­op­ment, a weak­ness she’s been work­ing on by re­cruit­ing top tal­ent from that oth­er side of the R&D di­men­sion. That strat­e­gy re­cent­ly led to the ar­rival of Mer­ck R&D chief Roger Perl­mut­ter on the board.

Koller told me:

The ear­ly days were very chal­leng­ing, I won’t lie. I was al­so ac­tu­al­ly new to the space…To build some­thing from scratch you need a net­work, you need to be con­nect­ed to peo­ple that you can re­cruit, you need to ask ad­vice. When I came in­to this I had none of that. So it was an in­cred­i­bly chal­leng­ing thing to just start from a blank slate in a space where you were a new­com­er and fig­ure out even how to go about build­ing a wet lab that I had nev­er built be­fore.

Now she’s work­ing with Har­g­reaves, one of the top neu­ro­sci­en­tists in the field, as Koller and her team con­tin­ue to build up the com­pa­ny.

About five years ago, Amit Etkin had a breakthrough.

The Stanford neurologist, a soft-spoken demi-prodigy who became a professor while still a resident, had been obsessed for a decade with how to better define psychiatric disorders. Drugs for depression or bipolar disorder didn’t work for many patients with the conditions, and he suspected the reason was how traditional diagnoses didn’t actually get at the heart of what was going on in a patient’s brain.

While falling behind the biggest PD-(L)1 drugs in terms of sales, Roche has looked to carve out a space for its Tecentriq with a growing expertise in lung cancer. The drug will now take an early lead in the sought-after adjuvant setting — but competitors are on the way.

The FDA on Friday approved Tecentriq as an adjuvant therapy for patients with Stage II-IIIA non small cell lung cancer with PD-(L)1 scores greater than or equal to 1, making it the first drug of its kind approved in an early setting that covers around 40% of all NSCLC patients.

The FDA’s Vaccines and Related Biological Products Advisory Committee (VRBPAC) on Friday voted 19-0 in favor of authorizing a second shot of J&J’s Covid-19 vaccine to follow at least two months after the initial dose.

Regulators don’t have to follow VRBPAC’s recommendation, but they almost always do. Considering that the CDC’s advisory committee has already been set to review the expanded EUA, VRBPAC’s recommendation is likely to be adopted.